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,
 275                      struct sw_flow_key *key,
 276                      bool post_ct)
 277 {
 278         ovs_ct_update_key(skb, NULL, key, post_ct, false);
 279 }
 280
 281 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 282                    const struct sw_flow_key *output, struct sk_buff *skb)
 283 {
 284         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 285                 return -EMSGSIZE;
 286
 287         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 288             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 289                 return -EMSGSIZE;
 290
 291         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 292             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 293                 return -EMSGSIZE;
 294
 295         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 296             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 297                     &output->ct.labels))
 298                 return -EMSGSIZE;
 299
 300         if (swkey->ct_orig_proto) {
 301                 if (swkey->eth.type == htons(ETH_P_IP)) {
 302                         struct ovs_key_ct_tuple_ipv4 orig;
 303
 304                         memset(&orig, 0, sizeof(orig));
 305                         orig.ipv4_src = output->ipv4.ct_orig.src;
 306                         orig.ipv4_dst = output->ipv4.ct_orig.dst;
 307                         orig.src_port = output->ct.orig_tp.src;
 308                         orig.dst_port = output->ct.orig_tp.dst;
 309                         orig.ipv4_proto = output->ct_orig_proto;
 310
 311                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 312                                     sizeof(orig), &orig))
 313                                 return -EMSGSIZE;
 314                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 315                         struct ovs_key_ct_tuple_ipv6 orig;
 316
 317                         memset(&orig, 0, sizeof(orig));
 318                         memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
 319                                sizeof(orig.ipv6_src));
 320                         memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
 321                                sizeof(orig.ipv6_dst));
 322                         orig.src_port = output->ct.orig_tp.src;
 323                         orig.dst_port = output->ct.orig_tp.dst;
 324                         orig.ipv6_proto = output->ct_orig_proto;
 325
 326                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 327                                     sizeof(orig), &orig))
 328                                 return -EMSGSIZE;
 329                 }
 330         }
 331
 332         return 0;
 333 }
 334
 335 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 336                            u32 ct_mark, u32 mask)
 337 {
 338 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 339         u32 new_mark;
 340
 341         new_mark = ct_mark | (ct->mark & ~(mask));
 342         if (ct->mark != new_mark) {
 343                 ct->mark = new_mark;
 344                 if (nf_ct_is_confirmed(ct))
 345                         nf_conntrack_event_cache(IPCT_MARK, ct);
 346                 key->ct.mark = new_mark;
 347         }
 348
 349         return 0;
 350 #else
 351         return -ENOTSUPP;
 352 #endif
 353 }
 354
 355 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 356 {
 357         struct nf_conn_labels *cl;
 358
 359         cl = nf_ct_labels_find(ct);
 360         if (!cl) {
 361                 nf_ct_labels_ext_add(ct);
 362                 cl = nf_ct_labels_find(ct);
 363         }
 364
 365         return cl;
 366 }
 367
 368 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 369  * since the new connection is not yet confirmed, and thus no-one else has
 370  * access to it's labels, we simply write them over.
 371  */
 372 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 373                               const struct ovs_key_ct_labels *labels,
 374                               const struct ovs_key_ct_labels *mask)
 375 {
 376         struct nf_conn_labels *cl, *master_cl;
 377         bool have_mask = labels_nonzero(mask);
 378
 379         /* Inherit master's labels to the related connection? */
 380         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 381
 382         if (!master_cl && !have_mask)
 383                 return 0;   /* Nothing to do. */
 384
 385         cl = ovs_ct_get_conn_labels(ct);
 386         if (!cl)
 387                 return -ENOSPC;
 388
 389         /* Inherit the master's labels, if any. */
 390         if (master_cl)
 391                 *cl = *master_cl;
 392
 393         if (have_mask) {
 394                 u32 *dst = (u32 *)cl->bits;
 395                 int i;
 396
 397                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 398                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 399                                 (labels->ct_labels_32[i]
 400                                  & mask->ct_labels_32[i]);
 401         }
 402
 403         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 404          * IPCT_LABEL bit is set in the event cache.
 405          */
 406         nf_conntrack_event_cache(IPCT_LABEL, ct);
 407
 408         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 409
 410         return 0;
 411 }
 412
 413 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 414                              const struct ovs_key_ct_labels *labels,
 415                              const struct ovs_key_ct_labels *mask)
 416 {
 417         struct nf_conn_labels *cl;
 418         int err;
 419
 420         cl = ovs_ct_get_conn_labels(ct);
 421         if (!cl)
 422                 return -ENOSPC;
 423
 424         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 425                                     mask->ct_labels_32,
 426                                     OVS_CT_LABELS_LEN_32);
 427         if (err)
 428                 return err;
 429
 430         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 431
 432         return 0;
 433 }
 434
 435 /* 'skb' should already be pulled to nh_ofs. */
 436 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 437 {
 438         const struct nf_conntrack_helper *helper;
 439         const struct nf_conn_help *help;
 440         enum ip_conntrack_info ctinfo;
 441         unsigned int protoff;
 442         struct nf_conn *ct;
 443         int err;
 444
 445         ct = nf_ct_get(skb, &ctinfo);
 446         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 447                 return NF_ACCEPT;
 448
 449         help = nfct_help(ct);
 450         if (!help)
 451                 return NF_ACCEPT;
 452
 453         helper = rcu_dereference(help->helper);
 454         if (!helper)
 455                 return NF_ACCEPT;
 456
 457         switch (proto) {
 458         case NFPROTO_IPV4:
 459                 protoff = ip_hdrlen(skb);
 460                 break;
 461         case NFPROTO_IPV6: {
 462                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 463                 __be16 frag_off;
 464                 int ofs;
 465
 466                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 467                                        &frag_off);
 468                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 469                         pr_debug("proto header not found\n");
 470                         return NF_ACCEPT;
 471                 }
 472                 protoff = ofs;
 473                 break;
 474         }
 475         default:
 476                 WARN_ONCE(1, "helper invoked on non-IP family!");
 477                 return NF_DROP;
 478         }
 479
 480         err = helper->help(skb, protoff, ct, ctinfo);
 481         if (err != NF_ACCEPT)
 482                 return err;
 483
 484         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 485          * FTP with NAT) adusting the TCP payload size when mangling IP
 486          * addresses and/or port numbers in the text-based control connection.
 487          */
 488         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 489             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 490                 return NF_DROP;
 491         return NF_ACCEPT;
 492 }
 493
 494 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 495  * value if 'skb' is freed.
 496  */
 497 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 498                             u16 zone, struct sk_buff *skb)
 499 {
 500         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 501         int err;
 502
 503         if (key->eth.type == htons(ETH_P_IP)) {
 504                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 505
 506                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 507                 err = ip_defrag(net, skb, user);
 508                 if (err)
 509                         return err;
 510
 511                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 512 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 513         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 514                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 515
 516                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 517                 err = nf_ct_frag6_gather(net, skb, user);
 518                 if (err) {
 519                         if (err != -EINPROGRESS)
 520                                 kfree_skb(skb);
 521                         return err;
 522                 }
 523
 524                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 525                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 526 #endif
 527         } else {
 528                 kfree_skb(skb);
 529                 return -EPFNOSUPPORT;
 530         }
 531
 532         /* The key extracted from the fragment that completed this datagram
 533          * likely didn't have an L4 header, so regenerate it.
 534          */
 535         ovs_flow_key_update_l3l4(skb, key);
 536
 537         key->ip.frag = OVS_FRAG_TYPE_NONE;
 538         skb_clear_hash(skb);
 539         skb->ignore_df = 1;
 540         *OVS_CB(skb) = ovs_cb;
 541
 542         return 0;
 543 }
 544
 545 static struct nf_conntrack_expect *
 546 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 547                    u16 proto, const struct sk_buff *skb)
 548 {
 549         struct nf_conntrack_tuple tuple;
 550         struct nf_conntrack_expect *exp;
 551
 552         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 553                 return NULL;
 554
 555         exp = __nf_ct_expect_find(net, zone, &tuple);
 556         if (exp) {
 557                 struct nf_conntrack_tuple_hash *h;
 558
 559                 /* Delete existing conntrack entry, if it clashes with the
 560                  * expectation.  This can happen since conntrack ALGs do not
 561                  * check for clashes between (new) expectations and existing
 562                  * conntrack entries.  nf_conntrack_in() will check the
 563                  * expectations only if a conntrack entry can not be found,
 564                  * which can lead to OVS finding the expectation (here) in the
 565                  * init direction, but which will not be removed by the
 566                  * nf_conntrack_in() call, if a matching conntrack entry is
 567                  * found instead.  In this case all init direction packets
 568                  * would be reported as new related packets, while reply
 569                  * direction packets would be reported as un-related
 570                  * established packets.
 571                  */
 572                 h = nf_conntrack_find_get(net, zone, &tuple);
 573                 if (h) {
 574                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 575
 576                         nf_ct_delete(ct, 0, 0);
 577                         nf_ct_put(ct);
 578                 }
 579         }
 580
 581         return exp;
 582 }
 583
 584 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 585 static enum ip_conntrack_info
 586 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 587 {
 588         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 589
 590         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 591                 return IP_CT_ESTABLISHED_REPLY;
 592         /* Once we've had two way comms, always ESTABLISHED. */
 593         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 594                 return IP_CT_ESTABLISHED;
 595         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 596                 return IP_CT_RELATED;
 597         return IP_CT_NEW;
 598 }
 599
 600 /* Find an existing connection which this packet belongs to without
 601  * re-attributing statistics or modifying the connection state.  This allows an
 602  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 603  *
 604  * Must be called with rcu_read_lock.
 605  *
 606  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 607  * if there is no existing entry.
 608  */
 609 static struct nf_conn *
 610 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 611                      u8 l3num, struct sk_buff *skb, bool natted)
 612 {
 613         struct nf_conntrack_tuple tuple;
 614         struct nf_conntrack_tuple_hash *h;
 615         struct nf_conn *ct;
 616
 617         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 618                                net, &tuple)) {
 619                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 620                 return NULL;
 621         }
 622
 623         /* Must invert the tuple if skb has been transformed by NAT. */
 624         if (natted) {
 625                 struct nf_conntrack_tuple inverse;
 626
 627                 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
 628                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 629                         return NULL;
 630                 }
 631                 tuple = inverse;
 632         }
 633
 634         /* look for tuple match */
 635         h = nf_conntrack_find_get(net, zone, &tuple);
 636         if (!h)
 637                 return NULL;   /* Not found. */
 638
 639         ct = nf_ct_tuplehash_to_ctrack(h);
 640
 641         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 642          * select the other tuplehash to get the right 'ctinfo' bits for this
 643          * packet.
 644          */
 645         if (natted)
 646                 h = &ct->tuplehash[!h->tuple.dst.dir];
 647
 648         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 649         return ct;
 650 }
 651
 652 static
 653 struct nf_conn *ovs_ct_executed(struct net *net,
 654                                 const struct sw_flow_key *key,
 655                                 const struct ovs_conntrack_info *info,
 656                                 struct sk_buff *skb,
 657                                 bool *ct_executed)
 658 {
 659         struct nf_conn *ct = NULL;
 660
 661         /* If no ct, check if we have evidence that an existing conntrack entry
 662          * might be found for this skb.  This happens when we lose a skb->_nfct
 663          * due to an upcall, or if the direction is being forced.  If the
 664          * connection was not confirmed, it is not cached and needs to be run
 665          * through conntrack again.
 666          */
 667         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 668                        !(key->ct_state & OVS_CS_F_INVALID) &&
 669                        (key->ct_zone == info->zone.id);
 670
 671         if (*ct_executed || (!key->ct_state && info->force)) {
 672                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 673                                           !!(key->ct_state &
 674                                           OVS_CS_F_NAT_MASK));
 675         }
 676
 677         return ct;
 678 }
 679
 680 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 681 static bool skb_nfct_cached(struct net *net,
 682                             const struct sw_flow_key *key,
 683                             const struct ovs_conntrack_info *info,
 684                             struct sk_buff *skb)
 685 {
 686         enum ip_conntrack_info ctinfo;
 687         struct nf_conn *ct;
 688         bool ct_executed = true;
 689
 690         ct = nf_ct_get(skb, &ctinfo);
 691         if (!ct)
 692                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 693
 694         if (ct)
 695                 nf_ct_get(skb, &ctinfo);
 696         else
 697                 return false;
 698
 699         if (!net_eq(net, read_pnet(&ct->ct_net)))
 700                 return false;
 701         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 702                 return false;
 703         if (info->helper) {
 704                 struct nf_conn_help *help;
 705
 706                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 707                 if (help && rcu_access_pointer(help->helper) != info->helper)
 708                         return false;
 709         }
 710         if (info->nf_ct_timeout) {
 711                 struct nf_conn_timeout *timeout_ext;
 712
 713                 timeout_ext = nf_ct_timeout_find(ct);
 714                 if (!timeout_ext || info->nf_ct_timeout !=
 715                     rcu_dereference(timeout_ext->timeout))
 716                         return false;
 717         }
 718         /* Force conntrack entry direction to the current packet? */
 719         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 720                 /* Delete the conntrack entry if confirmed, else just release
 721                  * the reference.
 722                  */
 723                 if (nf_ct_is_confirmed(ct))
 724                         nf_ct_delete(ct, 0, 0);
 725
 726                 nf_ct_put(ct);
 727                 nf_ct_set(skb, NULL, 0);
 728                 return false;
 729         }
 730
 731         return ct_executed;
 732 }
 733
 734 #if IS_ENABLED(CONFIG_NF_NAT)
 735 static void ovs_nat_update_key(struct sw_flow_key *key,
 736                                const struct sk_buff *skb,
 737                                enum nf_nat_manip_type maniptype)
 738 {
 739         if (maniptype == NF_NAT_MANIP_SRC) {
 740                 __be16 src;
 741
 742                 key->ct_state |= OVS_CS_F_SRC_NAT;
 743                 if (key->eth.type == htons(ETH_P_IP))
 744                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 745                 else if (key->eth.type == htons(ETH_P_IPV6))
 746                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 747                                sizeof(key->ipv6.addr.src));
 748                 else
 749                         return;
 750
 751                 if (key->ip.proto == IPPROTO_UDP)
 752                         src = udp_hdr(skb)->source;
 753                 else if (key->ip.proto == IPPROTO_TCP)
 754                         src = tcp_hdr(skb)->source;
 755                 else if (key->ip.proto == IPPROTO_SCTP)
 756                         src = sctp_hdr(skb)->source;
 757                 else
 758                         return;
 759
 760                 key->tp.src = src;
 761         } else {
 762                 __be16 dst;
 763
 764                 key->ct_state |= OVS_CS_F_DST_NAT;
 765                 if (key->eth.type == htons(ETH_P_IP))
 766                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 767                 else if (key->eth.type == htons(ETH_P_IPV6))
 768                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 769                                sizeof(key->ipv6.addr.dst));
 770                 else
 771                         return;
 772
 773                 if (key->ip.proto == IPPROTO_UDP)
 774                         dst = udp_hdr(skb)->dest;
 775                 else if (key->ip.proto == IPPROTO_TCP)
 776                         dst = tcp_hdr(skb)->dest;
 777                 else if (key->ip.proto == IPPROTO_SCTP)
 778                         dst = sctp_hdr(skb)->dest;
 779                 else
 780                         return;
 781
 782                 key->tp.dst = dst;
 783         }
 784 }
 785
 786 /* Modelled after nf_nat_ipv[46]_fn().
 787  * range is only used for new, uninitialized NAT state.
 788  * Returns either NF_ACCEPT or NF_DROP.
 789  */
 790 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 791                               enum ip_conntrack_info ctinfo,
 792                               const struct nf_nat_range2 *range,
 793                               enum nf_nat_manip_type maniptype, struct sw_flow_key *key)
 794 {
 795         int hooknum, nh_off, err = NF_ACCEPT;
 796
 797         nh_off = skb_network_offset(skb);
 798         skb_pull_rcsum(skb, nh_off);
 799
 800         /* See HOOK2MANIP(). */
 801         if (maniptype == NF_NAT_MANIP_SRC)
 802                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 803         else
 804                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 805
 806         switch (ctinfo) {
 807         case IP_CT_RELATED:
 808         case IP_CT_RELATED_REPLY:
 809                 if (IS_ENABLED(CONFIG_NF_NAT) &&
 810                     skb->protocol == htons(ETH_P_IP) &&
 811                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 812                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 813                                                            hooknum))
 814                                 err = NF_DROP;
 815                         goto push;
 816                 } else if (IS_ENABLED(CONFIG_IPV6) &&
 817                            skb->protocol == htons(ETH_P_IPV6)) {
 818                         __be16 frag_off;
 819                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 820                         int hdrlen = ipv6_skip_exthdr(skb,
 821                                                       sizeof(struct ipv6hdr),
 822                                                       &nexthdr, &frag_off);
 823
 824                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 825                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 826                                                                      ctinfo,
 827                                                                      hooknum,
 828                                                                      hdrlen))
 829                                         err = NF_DROP;
 830                                 goto push;
 831                         }
 832                 }
 833                 /* Non-ICMP, fall thru to initialize if needed. */
 834                 fallthrough;
 835         case IP_CT_NEW:
 836                 /* Seen it before?  This can happen for loopback, retrans,
 837                  * or local packets.
 838                  */
 839                 if (!nf_nat_initialized(ct, maniptype)) {
 840                         /* Initialize according to the NAT action. */
 841                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 842                                 /* Action is set up to establish a new
 843                                  * mapping.
 844                                  */
 845                                 ? nf_nat_setup_info(ct, range, maniptype)
 846                                 : nf_nat_alloc_null_binding(ct, hooknum);
 847                         if (err != NF_ACCEPT)
 848                                 goto push;
 849                 }
 850                 break;
 851
 852         case IP_CT_ESTABLISHED:
 853         case IP_CT_ESTABLISHED_REPLY:
 854                 break;
 855
 856         default:
 857                 err = NF_DROP;
 858                 goto push;
 859         }
 860
 861         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 862 push:
 863         skb_push_rcsum(skb, nh_off);
 864
 865         /* Update the flow key if NAT successful. */
 866         if (err == NF_ACCEPT)
 867                 ovs_nat_update_key(key, skb, maniptype);
 868
 869         return err;
 870 }
 871
 872 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 873 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 874                       const struct ovs_conntrack_info *info,
 875                       struct sk_buff *skb, struct nf_conn *ct,
 876                       enum ip_conntrack_info ctinfo)
 877 {
 878         enum nf_nat_manip_type maniptype;
 879         int err;
 880
 881         /* Add NAT extension if not confirmed yet. */
 882         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 883                 return NF_ACCEPT;   /* Can't NAT. */
 884
 885         /* Determine NAT type.
 886          * Check if the NAT type can be deduced from the tracked connection.
 887          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 888          * when committing.
 889          */
 890         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 891             ct->status & IPS_NAT_MASK &&
 892             (ctinfo != IP_CT_RELATED || info->commit)) {
 893                 /* NAT an established or related connection like before. */
 894                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 895                         /* This is the REPLY direction for a connection
 896                          * for which NAT was applied in the forward
 897                          * direction.  Do the reverse NAT.
 898                          */
 899                         maniptype = ct->status & IPS_SRC_NAT
 900                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 901                 else
 902                         maniptype = ct->status & IPS_SRC_NAT
 903                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 904         } else if (info->nat & OVS_CT_SRC_NAT) {
 905                 maniptype = NF_NAT_MANIP_SRC;
 906         } else if (info->nat & OVS_CT_DST_NAT) {
 907                 maniptype = NF_NAT_MANIP_DST;
 908         } else {
 909                 return NF_ACCEPT; /* Connection is not NATed. */
 910         }
 911         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype, key);
 912
 913         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
 914                 if (ct->status & IPS_SRC_NAT) {
 915                         if (maniptype == NF_NAT_MANIP_SRC)
 916                                 maniptype = NF_NAT_MANIP_DST;
 917                         else
 918                                 maniptype = NF_NAT_MANIP_SRC;
 919
 920                         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
 921                                                  maniptype, key);
 922                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 923                         err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
 924                                                  NF_NAT_MANIP_SRC, key);
 925                 }
 926         }
 927
 928         return err;
 929 }
 930 #else /* !CONFIG_NF_NAT */
 931 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 932                       const struct ovs_conntrack_info *info,
 933                       struct sk_buff *skb, struct nf_conn *ct,
 934                       enum ip_conntrack_info ctinfo)
 935 {
 936         return NF_ACCEPT;
 937 }
 938 #endif
 939
 940 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 941  * not done already.  Update key with new CT state after passing the packet
 942  * through conntrack.
 943  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 944  * set to NULL and 0 will be returned.
 945  */
 946 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 947                            const struct ovs_conntrack_info *info,
 948                            struct sk_buff *skb)
 949 {
 950         /* If we are recirculating packets to match on conntrack fields and
 951          * committing with a separate conntrack action,  then we don't need to
 952          * actually run the packet through conntrack twice unless it's for a
 953          * different zone.
 954          */
 955         bool cached = skb_nfct_cached(net, key, info, skb);
 956         enum ip_conntrack_info ctinfo;
 957         struct nf_conn *ct;
 958
 959         if (!cached) {
 960                 struct nf_hook_state state = {
 961                         .hook = NF_INET_PRE_ROUTING,
 962                         .pf = info->family,
 963                         .net = net,
 964                 };
 965                 struct nf_conn *tmpl = info->ct;
 966                 int err;
 967
 968                 /* Associate skb with specified zone. */
 969                 if (tmpl) {
 970                         ct = nf_ct_get(skb, &ctinfo);
 971                         nf_ct_put(ct);
 972                         nf_conntrack_get(&tmpl->ct_general);
 973                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 974                 }
 975
 976                 err = nf_conntrack_in(skb, &state);
 977                 if (err != NF_ACCEPT)
 978                         return -ENOENT;
 979
 980                 /* Clear CT state NAT flags to mark that we have not yet done
 981                  * NAT after the nf_conntrack_in() call.  We can actually clear
 982                  * the whole state, as it will be re-initialized below.
 983                  */
 984                 key->ct_state = 0;
 985
 986                 /* Update the key, but keep the NAT flags. */
 987                 ovs_ct_update_key(skb, info, key, true, true);
 988         }
 989
 990         ct = nf_ct_get(skb, &ctinfo);
 991         if (ct) {
 992                 bool add_helper = false;
 993
 994                 /* Packets starting a new connection must be NATted before the
 995                  * helper, so that the helper knows about the NAT.  We enforce
 996                  * this by delaying both NAT and helper calls for unconfirmed
 997                  * connections until the committing CT action.  For later
 998                  * packets NAT and Helper may be called in either order.
 999                  *
1000                  * NAT will be done only if the CT action has NAT, and only
1001                  * once per packet (per zone), as guarded by the NAT bits in
1002                  * the key->ct_state.
1003                  */
1004                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1005                     (nf_ct_is_confirmed(ct) || info->commit) &&
1006                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1007                         return -EINVAL;
1008                 }
1009
1010                 /* Userspace may decide to perform a ct lookup without a helper
1011                  * specified followed by a (recirculate and) commit with one,
1012                  * or attach a helper in a later commit.  Therefore, for
1013                  * connections which we will commit, we may need to attach
1014                  * the helper here.
1015                  */
1016                 if (info->commit && info->helper && !nfct_help(ct)) {
1017                         int err = __nf_ct_try_assign_helper(ct, info->ct,
1018                                                             GFP_ATOMIC);
1019                         if (err)
1020                                 return err;
1021                         add_helper = true;
1022
1023                         /* helper installed, add seqadj if NAT is required */
1024                         if (info->nat && !nfct_seqadj(ct)) {
1025                                 if (!nfct_seqadj_ext_add(ct))
1026                                         return -EINVAL;
1027                         }
1028                 }
1029
1030                 /* Call the helper only if:
1031                  * - nf_conntrack_in() was executed above ("!cached") or a
1032                  *   helper was just attached ("add_helper") for a confirmed
1033                  *   connection, or
1034                  * - When committing an unconfirmed connection.
1035                  */
1036                 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1037                                               info->commit) &&
1038                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1039                         return -EINVAL;
1040                 }
1041
1042                 if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1043                     nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1044                         /* Be liberal for tcp packets so that out-of-window
1045                          * packets are not marked invalid.
1046                          */
1047                         nf_ct_set_tcp_be_liberal(ct);
1048                 }
1049         }
1050
1051         return 0;
1052 }
1053
1054 /* Lookup connection and read fields into key. */
1055 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1056                          const struct ovs_conntrack_info *info,
1057                          struct sk_buff *skb)
1058 {
1059         struct nf_conntrack_expect *exp;
1060
1061         /* If we pass an expected packet through nf_conntrack_in() the
1062          * expectation is typically removed, but the packet could still be
1063          * lost in upcall processing.  To prevent this from happening we
1064          * perform an explicit expectation lookup.  Expected connections are
1065          * always new, and will be passed through conntrack only when they are
1066          * committed, as it is OK to remove the expectation at that time.
1067          */
1068         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1069         if (exp) {
1070                 u8 state;
1071
1072                 /* NOTE: New connections are NATted and Helped only when
1073                  * committed, so we are not calling into NAT here.
1074                  */
1075                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1076                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1077         } else {
1078                 struct nf_conn *ct;
1079                 int err;
1080
1081                 err = __ovs_ct_lookup(net, key, info, skb);
1082                 if (err)
1083                         return err;
1084
1085                 ct = (struct nf_conn *)skb_nfct(skb);
1086                 if (ct)
1087                         nf_ct_deliver_cached_events(ct);
1088         }
1089
1090         return 0;
1091 }
1092
1093 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1094 {
1095         size_t i;
1096
1097         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1098                 if (labels->ct_labels_32[i])
1099                         return true;
1100
1101         return false;
1102 }
1103
1104 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1105 static struct hlist_head *ct_limit_hash_bucket(
1106         const struct ovs_ct_limit_info *info, u16 zone)
1107 {
1108         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1109 }
1110
1111 /* Call with ovs_mutex */
1112 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1113                          struct ovs_ct_limit *new_ct_limit)
1114 {
1115         struct ovs_ct_limit *ct_limit;
1116         struct hlist_head *head;
1117
1118         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1119         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1120                 if (ct_limit->zone == new_ct_limit->zone) {
1121                         hlist_replace_rcu(&ct_limit->hlist_node,
1122                                           &new_ct_limit->hlist_node);
1123                         kfree_rcu(ct_limit, rcu);
1124                         return;
1125                 }
1126         }
1127
1128         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1129 }
1130
1131 /* Call with ovs_mutex */
1132 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1133 {
1134         struct ovs_ct_limit *ct_limit;
1135         struct hlist_head *head;
1136         struct hlist_node *n;
1137
1138         head = ct_limit_hash_bucket(info, zone);
1139         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1140                 if (ct_limit->zone == zone) {
1141                         hlist_del_rcu(&ct_limit->hlist_node);
1142                         kfree_rcu(ct_limit, rcu);
1143                         return;
1144                 }
1145         }
1146 }
1147
1148 /* Call with RCU read lock */
1149 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1150 {
1151         struct ovs_ct_limit *ct_limit;
1152         struct hlist_head *head;
1153
1154         head = ct_limit_hash_bucket(info, zone);
1155         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1156                 if (ct_limit->zone == zone)
1157                         return ct_limit->limit;
1158         }
1159
1160         return info->default_limit;
1161 }
1162
1163 static int ovs_ct_check_limit(struct net *net,
1164                               const struct ovs_conntrack_info *info,
1165                               const struct nf_conntrack_tuple *tuple)
1166 {
1167         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1168         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1169         u32 per_zone_limit, connections;
1170         u32 conncount_key;
1171
1172         conncount_key = info->zone.id;
1173
1174         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1175         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1176                 return 0;
1177
1178         connections = nf_conncount_count(net, ct_limit_info->data,
1179                                          &conncount_key, tuple, &info->zone);
1180         if (connections > per_zone_limit)
1181                 return -ENOMEM;
1182
1183         return 0;
1184 }
1185 #endif
1186
1187 /* Lookup connection and confirm if unconfirmed. */
1188 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1189                          const struct ovs_conntrack_info *info,
1190                          struct sk_buff *skb)
1191 {
1192         enum ip_conntrack_info ctinfo;
1193         struct nf_conn *ct;
1194         int err;
1195
1196         err = __ovs_ct_lookup(net, key, info, skb);
1197         if (err)
1198                 return err;
1199
1200         /* The connection could be invalid, in which case this is a no-op.*/
1201         ct = nf_ct_get(skb, &ctinfo);
1202         if (!ct)
1203                 return 0;
1204
1205 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1206         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1207                 if (!nf_ct_is_confirmed(ct)) {
1208                         err = ovs_ct_check_limit(net, info,
1209                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1210                         if (err) {
1211                                 net_warn_ratelimited("openvswitch: zone: %u "
1212                                         "exceeds conntrack limit\n",
1213                                         info->zone.id);
1214                                 return err;
1215                         }
1216                 }
1217         }
1218 #endif
1219
1220         /* Set the conntrack event mask if given.  NEW and DELETE events have
1221          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1222          * typically would receive many kinds of updates.  Setting the event
1223          * mask allows those events to be filtered.  The set event mask will
1224          * remain in effect for the lifetime of the connection unless changed
1225          * by a further CT action with both the commit flag and the eventmask
1226          * option. */
1227         if (info->have_eventmask) {
1228                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1229
1230                 if (cache)
1231                         cache->ctmask = info->eventmask;
1232         }
1233
1234         /* Apply changes before confirming the connection so that the initial
1235          * conntrack NEW netlink event carries the values given in the CT
1236          * action.
1237          */
1238         if (info->mark.mask) {
1239                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1240                                       info->mark.mask);
1241                 if (err)
1242                         return err;
1243         }
1244         if (!nf_ct_is_confirmed(ct)) {
1245                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1246                                          &info->labels.mask);
1247                 if (err)
1248                         return err;
1249         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1250                    labels_nonzero(&info->labels.mask)) {
1251                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1252                                         &info->labels.mask);
1253                 if (err)
1254                         return err;
1255         }
1256         /* This will take care of sending queued events even if the connection
1257          * is already confirmed.
1258          */
1259         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1260                 return -EINVAL;
1261
1262         return 0;
1263 }
1264
1265 /* Trim the skb to the length specified by the IP/IPv6 header,
1266  * removing any trailing lower-layer padding. This prepares the skb
1267  * for higher-layer processing that assumes skb->len excludes padding
1268  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1269  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1270  */
1271 static int ovs_skb_network_trim(struct sk_buff *skb)
1272 {
1273         unsigned int len;
1274         int err;
1275
1276         switch (skb->protocol) {
1277         case htons(ETH_P_IP):
1278                 len = ntohs(ip_hdr(skb)->tot_len);
1279                 break;
1280         case htons(ETH_P_IPV6):
1281                 len = sizeof(struct ipv6hdr)
1282                         + ntohs(ipv6_hdr(skb)->payload_len);
1283                 break;
1284         default:
1285                 len = skb->len;
1286         }
1287
1288         err = pskb_trim_rcsum(skb, len);
1289         if (err)
1290                 kfree_skb(skb);
1291
1292         return err;
1293 }
1294
1295 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1296  * value if 'skb' is freed.
1297  */
1298 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1299                    struct sw_flow_key *key,
1300                    const struct ovs_conntrack_info *info)
1301 {
1302         int nh_ofs;
1303         int err;
1304
1305         /* The conntrack module expects to be working at L3. */
1306         nh_ofs = skb_network_offset(skb);
1307         skb_pull_rcsum(skb, nh_ofs);
1308
1309         err = ovs_skb_network_trim(skb);
1310         if (err)
1311                 return err;
1312
1313         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1314                 err = handle_fragments(net, key, info->zone.id, skb);
1315                 if (err)
1316                         return err;
1317         }
1318
1319         if (info->commit)
1320                 err = ovs_ct_commit(net, key, info, skb);
1321         else
1322                 err = ovs_ct_lookup(net, key, info, skb);
1323
1324         skb_push_rcsum(skb, 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         enum ip_conntrack_info ctinfo;
1333         struct nf_conn *ct;
1334
1335         ct = nf_ct_get(skb, &ctinfo);
1336
1337         nf_ct_put(ct);
1338         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1339         ovs_ct_fill_key(skb, key, false);
1340
1341         return 0;
1342 }
1343
1344 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1345                              const struct sw_flow_key *key, bool log)
1346 {
1347         struct nf_conntrack_helper *helper;
1348         struct nf_conn_help *help;
1349         int ret = 0;
1350
1351         helper = nf_conntrack_helper_try_module_get(name, info->family,
1352                                                     key->ip.proto);
1353         if (!helper) {
1354                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1355                 return -EINVAL;
1356         }
1357
1358         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1359         if (!help) {
1360                 nf_conntrack_helper_put(helper);
1361                 return -ENOMEM;
1362         }
1363
1364 #if IS_ENABLED(CONFIG_NF_NAT)
1365         if (info->nat) {
1366                 ret = nf_nat_helper_try_module_get(name, info->family,
1367                                                    key->ip.proto);
1368                 if (ret) {
1369                         nf_conntrack_helper_put(helper);
1370                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1371                                   name, ret);
1372                         return ret;
1373                 }
1374         }
1375 #endif
1376         rcu_assign_pointer(help->helper, helper);
1377         info->helper = helper;
1378         return ret;
1379 }
1380
1381 #if IS_ENABLED(CONFIG_NF_NAT)
1382 static int parse_nat(const struct nlattr *attr,
1383                      struct ovs_conntrack_info *info, bool log)
1384 {
1385         struct nlattr *a;
1386         int rem;
1387         bool have_ip_max = false;
1388         bool have_proto_max = false;
1389         bool ip_vers = (info->family == NFPROTO_IPV6);
1390
1391         nla_for_each_nested(a, attr, rem) {
1392                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1393                         [OVS_NAT_ATTR_SRC] = {0, 0},
1394                         [OVS_NAT_ATTR_DST] = {0, 0},
1395                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1396                                                  sizeof(struct in6_addr)},
1397                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1398                                                  sizeof(struct in6_addr)},
1399                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1400                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1401                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1402                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1403                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1404                 };
1405                 int type = nla_type(a);
1406
1407                 if (type > OVS_NAT_ATTR_MAX) {
1408                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1409                                   type, OVS_NAT_ATTR_MAX);
1410                         return -EINVAL;
1411                 }
1412
1413                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1414                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1415                                   type, nla_len(a),
1416                                   ovs_nat_attr_lens[type][ip_vers]);
1417                         return -EINVAL;
1418                 }
1419
1420                 switch (type) {
1421                 case OVS_NAT_ATTR_SRC:
1422                 case OVS_NAT_ATTR_DST:
1423                         if (info->nat) {
1424                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1425                                 return -ERANGE;
1426                         }
1427                         info->nat |= OVS_CT_NAT;
1428                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1429                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1430                         break;
1431
1432                 case OVS_NAT_ATTR_IP_MIN:
1433                         nla_memcpy(&info->range.min_addr, a,
1434                                    sizeof(info->range.min_addr));
1435                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1436                         break;
1437
1438                 case OVS_NAT_ATTR_IP_MAX:
1439                         have_ip_max = true;
1440                         nla_memcpy(&info->range.max_addr, a,
1441                                    sizeof(info->range.max_addr));
1442                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1443                         break;
1444
1445                 case OVS_NAT_ATTR_PROTO_MIN:
1446                         info->range.min_proto.all = htons(nla_get_u16(a));
1447                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1448                         break;
1449
1450                 case OVS_NAT_ATTR_PROTO_MAX:
1451                         have_proto_max = true;
1452                         info->range.max_proto.all = htons(nla_get_u16(a));
1453                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1454                         break;
1455
1456                 case OVS_NAT_ATTR_PERSISTENT:
1457                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1458                         break;
1459
1460                 case OVS_NAT_ATTR_PROTO_HASH:
1461                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1462                         break;
1463
1464                 case OVS_NAT_ATTR_PROTO_RANDOM:
1465                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1466                         break;
1467
1468                 default:
1469                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1470                         return -EINVAL;
1471                 }
1472         }
1473
1474         if (rem > 0) {
1475                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1476                 return -EINVAL;
1477         }
1478         if (!info->nat) {
1479                 /* Do not allow flags if no type is given. */
1480                 if (info->range.flags) {
1481                         OVS_NLERR(log,
1482                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1483                                   );
1484                         return -EINVAL;
1485                 }
1486                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1487         } else if (!info->commit) {
1488                 OVS_NLERR(log,
1489                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1490                           );
1491                 return -EINVAL;
1492         }
1493         /* Allow missing IP_MAX. */
1494         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1495                 memcpy(&info->range.max_addr, &info->range.min_addr,
1496                        sizeof(info->range.max_addr));
1497         }
1498         /* Allow missing PROTO_MAX. */
1499         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1500             !have_proto_max) {
1501                 info->range.max_proto.all = info->range.min_proto.all;
1502         }
1503         return 0;
1504 }
1505 #endif
1506
1507 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1508         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1509         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1510         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1511                                     .maxlen = sizeof(u16) },
1512         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1513                                     .maxlen = sizeof(struct md_mark) },
1514         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1515                                     .maxlen = sizeof(struct md_labels) },
1516         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1517                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1518 #if IS_ENABLED(CONFIG_NF_NAT)
1519         /* NAT length is checked when parsing the nested attributes. */
1520         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1521 #endif
1522         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1523                                     .maxlen = sizeof(u32) },
1524         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1525                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1526 };
1527
1528 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1529                     const char **helper, bool log)
1530 {
1531         struct nlattr *a;
1532         int rem;
1533
1534         nla_for_each_nested(a, attr, rem) {
1535                 int type = nla_type(a);
1536                 int maxlen;
1537                 int minlen;
1538
1539                 if (type > OVS_CT_ATTR_MAX) {
1540                         OVS_NLERR(log,
1541                                   "Unknown conntrack attr (type=%d, max=%d)",
1542                                   type, OVS_CT_ATTR_MAX);
1543                         return -EINVAL;
1544                 }
1545
1546                 maxlen = ovs_ct_attr_lens[type].maxlen;
1547                 minlen = ovs_ct_attr_lens[type].minlen;
1548                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1549                         OVS_NLERR(log,
1550                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1551                                   type, nla_len(a), maxlen);
1552                         return -EINVAL;
1553                 }
1554
1555                 switch (type) {
1556                 case OVS_CT_ATTR_FORCE_COMMIT:
1557                         info->force = true;
1558                         fallthrough;
1559                 case OVS_CT_ATTR_COMMIT:
1560                         info->commit = true;
1561                         break;
1562 #ifdef CONFIG_NF_CONNTRACK_ZONES
1563                 case OVS_CT_ATTR_ZONE:
1564                         info->zone.id = nla_get_u16(a);
1565                         break;
1566 #endif
1567 #ifdef CONFIG_NF_CONNTRACK_MARK
1568                 case OVS_CT_ATTR_MARK: {
1569                         struct md_mark *mark = nla_data(a);
1570
1571                         if (!mark->mask) {
1572                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1573                                 return -EINVAL;
1574                         }
1575                         info->mark = *mark;
1576                         break;
1577                 }
1578 #endif
1579 #ifdef CONFIG_NF_CONNTRACK_LABELS
1580                 case OVS_CT_ATTR_LABELS: {
1581                         struct md_labels *labels = nla_data(a);
1582
1583                         if (!labels_nonzero(&labels->mask)) {
1584                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1585                                 return -EINVAL;
1586                         }
1587                         info->labels = *labels;
1588                         break;
1589                 }
1590 #endif
1591                 case OVS_CT_ATTR_HELPER:
1592                         *helper = nla_data(a);
1593                         if (!memchr(*helper, '\0', nla_len(a))) {
1594                                 OVS_NLERR(log, "Invalid conntrack helper");
1595                                 return -EINVAL;
1596                         }
1597                         break;
1598 #if IS_ENABLED(CONFIG_NF_NAT)
1599                 case OVS_CT_ATTR_NAT: {
1600                         int err = parse_nat(a, info, log);
1601
1602                         if (err)
1603                                 return err;
1604                         break;
1605                 }
1606 #endif
1607                 case OVS_CT_ATTR_EVENTMASK:
1608                         info->have_eventmask = true;
1609                         info->eventmask = nla_get_u32(a);
1610                         break;
1611 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1612                 case OVS_CT_ATTR_TIMEOUT:
1613                         memcpy(info->timeout, nla_data(a), nla_len(a));
1614                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1615                                 OVS_NLERR(log, "Invalid conntrack timeout");
1616                                 return -EINVAL;
1617                         }
1618                         break;
1619 #endif
1620
1621                 default:
1622                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1623                                   type);
1624                         return -EINVAL;
1625                 }
1626         }
1627
1628 #ifdef CONFIG_NF_CONNTRACK_MARK
1629         if (!info->commit && info->mark.mask) {
1630                 OVS_NLERR(log,
1631                           "Setting conntrack mark requires 'commit' flag.");
1632                 return -EINVAL;
1633         }
1634 #endif
1635 #ifdef CONFIG_NF_CONNTRACK_LABELS
1636         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1637                 OVS_NLERR(log,
1638                           "Setting conntrack labels requires 'commit' flag.");
1639                 return -EINVAL;
1640         }
1641 #endif
1642         if (rem > 0) {
1643                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1644                 return -EINVAL;
1645         }
1646
1647         return 0;
1648 }
1649
1650 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1651 {
1652         if (attr == OVS_KEY_ATTR_CT_STATE)
1653                 return true;
1654         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1655             attr == OVS_KEY_ATTR_CT_ZONE)
1656                 return true;
1657         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1658             attr == OVS_KEY_ATTR_CT_MARK)
1659                 return true;
1660         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1661             attr == OVS_KEY_ATTR_CT_LABELS) {
1662                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1663
1664                 return ovs_net->xt_label;
1665         }
1666
1667         return false;
1668 }
1669
1670 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1671                        const struct sw_flow_key *key,
1672                        struct sw_flow_actions **sfa,  bool log)
1673 {
1674         struct ovs_conntrack_info ct_info;
1675         const char *helper = NULL;
1676         u16 family;
1677         int err;
1678
1679         family = key_to_nfproto(key);
1680         if (family == NFPROTO_UNSPEC) {
1681                 OVS_NLERR(log, "ct family unspecified");
1682                 return -EINVAL;
1683         }
1684
1685         memset(&ct_info, 0, sizeof(ct_info));
1686         ct_info.family = family;
1687
1688         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1689                         NF_CT_DEFAULT_ZONE_DIR, 0);
1690
1691         err = parse_ct(attr, &ct_info, &helper, log);
1692         if (err)
1693                 return err;
1694
1695         /* Set up template for tracking connections in specific zones. */
1696         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1697         if (!ct_info.ct) {
1698                 OVS_NLERR(log, "Failed to allocate conntrack template");
1699                 return -ENOMEM;
1700         }
1701
1702         if (ct_info.timeout[0]) {
1703                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1704                                       ct_info.timeout))
1705                         pr_info_ratelimited("Failed to associated timeout "
1706                                             "policy `%s'\n", ct_info.timeout);
1707                 else
1708                         ct_info.nf_ct_timeout = rcu_dereference(
1709                                 nf_ct_timeout_find(ct_info.ct)->timeout);
1710
1711         }
1712
1713         if (helper) {
1714                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1715                 if (err)
1716                         goto err_free_ct;
1717         }
1718
1719         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1720                                  sizeof(ct_info), log);
1721         if (err)
1722                 goto err_free_ct;
1723
1724         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1725         return 0;
1726 err_free_ct:
1727         __ovs_ct_free_action(&ct_info);
1728         return err;
1729 }
1730
1731 #if IS_ENABLED(CONFIG_NF_NAT)
1732 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1733                                struct sk_buff *skb)
1734 {
1735         struct nlattr *start;
1736
1737         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1738         if (!start)
1739                 return false;
1740
1741         if (info->nat & OVS_CT_SRC_NAT) {
1742                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1743                         return false;
1744         } else if (info->nat & OVS_CT_DST_NAT) {
1745                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1746                         return false;
1747         } else {
1748                 goto out;
1749         }
1750
1751         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1752                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1753                     info->family == NFPROTO_IPV4) {
1754                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1755                                             info->range.min_addr.ip) ||
1756                             (info->range.max_addr.ip
1757                              != info->range.min_addr.ip &&
1758                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1759                                               info->range.max_addr.ip))))
1760                                 return false;
1761                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1762                            info->family == NFPROTO_IPV6) {
1763                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1764                                              &info->range.min_addr.in6) ||
1765                             (memcmp(&info->range.max_addr.in6,
1766                                     &info->range.min_addr.in6,
1767                                     sizeof(info->range.max_addr.in6)) &&
1768                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1769                                                &info->range.max_addr.in6))))
1770                                 return false;
1771                 } else {
1772                         return false;
1773                 }
1774         }
1775         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1776             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1777                          ntohs(info->range.min_proto.all)) ||
1778              (info->range.max_proto.all != info->range.min_proto.all &&
1779               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1780                           ntohs(info->range.max_proto.all)))))
1781                 return false;
1782
1783         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1784             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1785                 return false;
1786         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1787             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1788                 return false;
1789         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1790             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1791                 return false;
1792 out:
1793         nla_nest_end(skb, start);
1794
1795         return true;
1796 }
1797 #endif
1798
1799 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1800                           struct sk_buff *skb)
1801 {
1802         struct nlattr *start;
1803
1804         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1805         if (!start)
1806                 return -EMSGSIZE;
1807
1808         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1809                                             ? OVS_CT_ATTR_FORCE_COMMIT
1810                                             : OVS_CT_ATTR_COMMIT))
1811                 return -EMSGSIZE;
1812         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1813             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1814                 return -EMSGSIZE;
1815         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1816             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1817                     &ct_info->mark))
1818                 return -EMSGSIZE;
1819         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1820             labels_nonzero(&ct_info->labels.mask) &&
1821             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1822                     &ct_info->labels))
1823                 return -EMSGSIZE;
1824         if (ct_info->helper) {
1825                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1826                                    ct_info->helper->name))
1827                         return -EMSGSIZE;
1828         }
1829         if (ct_info->have_eventmask &&
1830             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1831                 return -EMSGSIZE;
1832         if (ct_info->timeout[0]) {
1833                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1834                         return -EMSGSIZE;
1835         }
1836
1837 #if IS_ENABLED(CONFIG_NF_NAT)
1838         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1839                 return -EMSGSIZE;
1840 #endif
1841         nla_nest_end(skb, start);
1842
1843         return 0;
1844 }
1845
1846 void ovs_ct_free_action(const struct nlattr *a)
1847 {
1848         struct ovs_conntrack_info *ct_info = nla_data(a);
1849
1850         __ovs_ct_free_action(ct_info);
1851 }
1852
1853 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1854 {
1855         if (ct_info->helper) {
1856 #if IS_ENABLED(CONFIG_NF_NAT)
1857                 if (ct_info->nat)
1858                         nf_nat_helper_put(ct_info->helper);
1859 #endif
1860                 nf_conntrack_helper_put(ct_info->helper);
1861         }
1862         if (ct_info->ct) {
1863                 if (ct_info->timeout[0])
1864                         nf_ct_destroy_timeout(ct_info->ct);
1865                 nf_ct_tmpl_free(ct_info->ct);
1866         }
1867 }
1868
1869 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1870 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1871 {
1872         int i, err;
1873
1874         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1875                                          GFP_KERNEL);
1876         if (!ovs_net->ct_limit_info)
1877                 return -ENOMEM;
1878
1879         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1880         ovs_net->ct_limit_info->limits =
1881                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1882                               GFP_KERNEL);
1883         if (!ovs_net->ct_limit_info->limits) {
1884                 kfree(ovs_net->ct_limit_info);
1885                 return -ENOMEM;
1886         }
1887
1888         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1889                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1890
1891         ovs_net->ct_limit_info->data =
1892                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1893
1894         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1895                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1896                 kfree(ovs_net->ct_limit_info->limits);
1897                 kfree(ovs_net->ct_limit_info);
1898                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1899                 return err;
1900         }
1901         return 0;
1902 }
1903
1904 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1905 {
1906         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1907         int i;
1908
1909         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1910         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1911                 struct hlist_head *head = &info->limits[i];
1912                 struct ovs_ct_limit *ct_limit;
1913
1914                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1915                                          lockdep_ovsl_is_held())
1916                         kfree_rcu(ct_limit, rcu);
1917         }
1918         kfree(info->limits);
1919         kfree(info);
1920 }
1921
1922 static struct sk_buff *
1923 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1924                              struct ovs_header **ovs_reply_header)
1925 {
1926         struct ovs_header *ovs_header = info->userhdr;
1927         struct sk_buff *skb;
1928
1929         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1930         if (!skb)
1931                 return ERR_PTR(-ENOMEM);
1932
1933         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1934                                         info->snd_seq,
1935                                         &dp_ct_limit_genl_family, 0, cmd);
1936
1937         if (!*ovs_reply_header) {
1938                 nlmsg_free(skb);
1939                 return ERR_PTR(-EMSGSIZE);
1940         }
1941         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1942
1943         return skb;
1944 }
1945
1946 static bool check_zone_id(int zone_id, u16 *pzone)
1947 {
1948         if (zone_id >= 0 && zone_id <= 65535) {
1949                 *pzone = (u16)zone_id;
1950                 return true;
1951         }
1952         return false;
1953 }
1954
1955 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1956                                        struct ovs_ct_limit_info *info)
1957 {
1958         struct ovs_zone_limit *zone_limit;
1959         int rem;
1960         u16 zone;
1961
1962         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1963         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1964
1965         while (rem >= sizeof(*zone_limit)) {
1966                 if (unlikely(zone_limit->zone_id ==
1967                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1968                         ovs_lock();
1969                         info->default_limit = zone_limit->limit;
1970                         ovs_unlock();
1971                 } else if (unlikely(!check_zone_id(
1972                                 zone_limit->zone_id, &zone))) {
1973                         OVS_NLERR(true, "zone id is out of range");
1974                 } else {
1975                         struct ovs_ct_limit *ct_limit;
1976
1977                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1978                         if (!ct_limit)
1979                                 return -ENOMEM;
1980
1981                         ct_limit->zone = zone;
1982                         ct_limit->limit = zone_limit->limit;
1983
1984                         ovs_lock();
1985                         ct_limit_set(info, ct_limit);
1986                         ovs_unlock();
1987                 }
1988                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1989                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1990                                 NLA_ALIGN(sizeof(*zone_limit)));
1991         }
1992
1993         if (rem)
1994                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1995
1996         return 0;
1997 }
1998
1999 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
2000                                        struct ovs_ct_limit_info *info)
2001 {
2002         struct ovs_zone_limit *zone_limit;
2003         int rem;
2004         u16 zone;
2005
2006         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2007         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2008
2009         while (rem >= sizeof(*zone_limit)) {
2010                 if (unlikely(zone_limit->zone_id ==
2011                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2012                         ovs_lock();
2013                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
2014                         ovs_unlock();
2015                 } else if (unlikely(!check_zone_id(
2016                                 zone_limit->zone_id, &zone))) {
2017                         OVS_NLERR(true, "zone id is out of range");
2018                 } else {
2019                         ovs_lock();
2020                         ct_limit_del(info, zone);
2021                         ovs_unlock();
2022                 }
2023                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2024                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2025                                 NLA_ALIGN(sizeof(*zone_limit)));
2026         }
2027
2028         if (rem)
2029                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2030
2031         return 0;
2032 }
2033
2034 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2035                                           struct sk_buff *reply)
2036 {
2037         struct ovs_zone_limit zone_limit = {
2038                 .zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2039                 .limit   = info->default_limit,
2040         };
2041
2042         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2043 }
2044
2045 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2046                                          struct nf_conncount_data *data,
2047                                          u16 zone_id, u32 limit,
2048                                          struct sk_buff *reply)
2049 {
2050         struct nf_conntrack_zone ct_zone;
2051         struct ovs_zone_limit zone_limit;
2052         u32 conncount_key = zone_id;
2053
2054         zone_limit.zone_id = zone_id;
2055         zone_limit.limit = limit;
2056         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2057
2058         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2059                                               &ct_zone);
2060         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2061 }
2062
2063 static int ovs_ct_limit_get_zone_limit(struct net *net,
2064                                        struct nlattr *nla_zone_limit,
2065                                        struct ovs_ct_limit_info *info,
2066                                        struct sk_buff *reply)
2067 {
2068         struct ovs_zone_limit *zone_limit;
2069         int rem, err;
2070         u32 limit;
2071         u16 zone;
2072
2073         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2074         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2075
2076         while (rem >= sizeof(*zone_limit)) {
2077                 if (unlikely(zone_limit->zone_id ==
2078                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2079                         err = ovs_ct_limit_get_default_limit(info, reply);
2080                         if (err)
2081                                 return err;
2082                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2083                                                         &zone))) {
2084                         OVS_NLERR(true, "zone id is out of range");
2085                 } else {
2086                         rcu_read_lock();
2087                         limit = ct_limit_get(info, zone);
2088                         rcu_read_unlock();
2089
2090                         err = __ovs_ct_limit_get_zone_limit(
2091                                 net, info->data, zone, limit, reply);
2092                         if (err)
2093                                 return err;
2094                 }
2095                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2096                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2097                                 NLA_ALIGN(sizeof(*zone_limit)));
2098         }
2099
2100         if (rem)
2101                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2102
2103         return 0;
2104 }
2105
2106 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2107                                            struct ovs_ct_limit_info *info,
2108                                            struct sk_buff *reply)
2109 {
2110         struct ovs_ct_limit *ct_limit;
2111         struct hlist_head *head;
2112         int i, err = 0;
2113
2114         err = ovs_ct_limit_get_default_limit(info, reply);
2115         if (err)
2116                 return err;
2117
2118         rcu_read_lock();
2119         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2120                 head = &info->limits[i];
2121                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2122                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2123                                 ct_limit->zone, ct_limit->limit, reply);
2124                         if (err)
2125                                 goto exit_err;
2126                 }
2127         }
2128
2129 exit_err:
2130         rcu_read_unlock();
2131         return err;
2132 }
2133
2134 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2135 {
2136         struct nlattr **a = info->attrs;
2137         struct sk_buff *reply;
2138         struct ovs_header *ovs_reply_header;
2139         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2140         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2141         int err;
2142
2143         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2144                                              &ovs_reply_header);
2145         if (IS_ERR(reply))
2146                 return PTR_ERR(reply);
2147
2148         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2149                 err = -EINVAL;
2150                 goto exit_err;
2151         }
2152
2153         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2154                                           ct_limit_info);
2155         if (err)
2156                 goto exit_err;
2157
2158         static_branch_enable(&ovs_ct_limit_enabled);
2159
2160         genlmsg_end(reply, ovs_reply_header);
2161         return genlmsg_reply(reply, info);
2162
2163 exit_err:
2164         nlmsg_free(reply);
2165         return err;
2166 }
2167
2168 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2169 {
2170         struct nlattr **a = info->attrs;
2171         struct sk_buff *reply;
2172         struct ovs_header *ovs_reply_header;
2173         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2174         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2175         int err;
2176
2177         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2178                                              &ovs_reply_header);
2179         if (IS_ERR(reply))
2180                 return PTR_ERR(reply);
2181
2182         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2183                 err = -EINVAL;
2184                 goto exit_err;
2185         }
2186
2187         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2188                                           ct_limit_info);
2189         if (err)
2190                 goto exit_err;
2191
2192         genlmsg_end(reply, ovs_reply_header);
2193         return genlmsg_reply(reply, info);
2194
2195 exit_err:
2196         nlmsg_free(reply);
2197         return err;
2198 }
2199
2200 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2201 {
2202         struct nlattr **a = info->attrs;
2203         struct nlattr *nla_reply;
2204         struct sk_buff *reply;
2205         struct ovs_header *ovs_reply_header;
2206         struct net *net = sock_net(skb->sk);
2207         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2208         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2209         int err;
2210
2211         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2212                                              &ovs_reply_header);
2213         if (IS_ERR(reply))
2214                 return PTR_ERR(reply);
2215
2216         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2217         if (!nla_reply) {
2218                 err = -EMSGSIZE;
2219                 goto exit_err;
2220         }
2221
2222         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2223                 err = ovs_ct_limit_get_zone_limit(
2224                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2225                         reply);
2226                 if (err)
2227                         goto exit_err;
2228         } else {
2229                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2230                                                       reply);
2231                 if (err)
2232                         goto exit_err;
2233         }
2234
2235         nla_nest_end(reply, nla_reply);
2236         genlmsg_end(reply, ovs_reply_header);
2237         return genlmsg_reply(reply, info);
2238
2239 exit_err:
2240         nlmsg_free(reply);
2241         return err;
2242 }
2243
2244 static const struct genl_small_ops ct_limit_genl_ops[] = {
2245         { .cmd = OVS_CT_LIMIT_CMD_SET,
2246                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2247                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2248                                            * privilege. */
2249                 .doit = ovs_ct_limit_cmd_set,
2250         },
2251         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2252                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2253                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2254                                            * privilege. */
2255                 .doit = ovs_ct_limit_cmd_del,
2256         },
2257         { .cmd = OVS_CT_LIMIT_CMD_GET,
2258                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2259                 .flags = 0,               /* OK for unprivileged users. */
2260                 .doit = ovs_ct_limit_cmd_get,
2261         },
2262 };
2263
2264 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2265         .name = OVS_CT_LIMIT_MCGROUP,
2266 };
2267
2268 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2269         .hdrsize = sizeof(struct ovs_header),
2270         .name = OVS_CT_LIMIT_FAMILY,
2271         .version = OVS_CT_LIMIT_VERSION,
2272         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2273         .policy = ct_limit_policy,
2274         .netnsok = true,
2275         .parallel_ops = true,
2276         .small_ops = ct_limit_genl_ops,
2277         .n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2278         .mcgrps = &ovs_ct_limit_multicast_group,
2279         .n_mcgrps = 1,
2280         .module = THIS_MODULE,
2281 };
2282 #endif
2283
2284 int ovs_ct_init(struct net *net)
2285 {
2286         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2287         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2288
2289         if (nf_connlabels_get(net, n_bits - 1)) {
2290                 ovs_net->xt_label = false;
2291                 OVS_NLERR(true, "Failed to set connlabel length");
2292         } else {
2293                 ovs_net->xt_label = true;
2294         }
2295
2296 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2297         return ovs_ct_limit_init(net, ovs_net);
2298 #else
2299         return 0;
2300 #endif
2301 }
2302
2303 void ovs_ct_exit(struct net *net)
2304 {
2305         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2306
2307 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2308         ovs_ct_limit_exit(net, ovs_net);
2309 #endif
2310
2311         if (ovs_net->xt_label)
2312                 nf_connlabels_put(net);
2313 }