net/openvswitch/conntrack.c

   1 /*
   2  * Copyright (c) 2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/openvswitch.h>
  16 #include <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <linux/sctp.h>
  19 #include <linux/static_key.h>
  20 #include <net/ip.h>
  21 #include <net/genetlink.h>
  22 #include <net/netfilter/nf_conntrack_core.h>
  23 #include <net/netfilter/nf_conntrack_count.h>
  24 #include <net/netfilter/nf_conntrack_helper.h>
  25 #include <net/netfilter/nf_conntrack_labels.h>
  26 #include <net/netfilter/nf_conntrack_seqadj.h>
  27 #include <net/netfilter/nf_conntrack_zones.h>
  28 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  29
  30 #ifdef CONFIG_NF_NAT_NEEDED
  31 #include <linux/netfilter/nf_nat.h>
  32 #include <net/netfilter/nf_nat_core.h>
  33 #include <net/netfilter/nf_nat_l3proto.h>
  34 #endif
  35
  36 #include "datapath.h"
  37 #include "conntrack.h"
  38 #include "flow.h"
  39 #include "flow_netlink.h"
  40
  41 struct ovs_ct_len_tbl {
  42         int maxlen;
  43         int minlen;
  44 };
  45
  46 /* Metadata mark for masked write to conntrack mark */
  47 struct md_mark {
  48         u32 value;
  49         u32 mask;
  50 };
  51
  52 /* Metadata label for masked write to conntrack label. */
  53 struct md_labels {
  54         struct ovs_key_ct_labels value;
  55         struct ovs_key_ct_labels mask;
  56 };
  57
  58 enum ovs_ct_nat {
  59         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  60         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  61         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  62 };
  63
  64 /* Conntrack action context for execution. */
  65 struct ovs_conntrack_info {
  66         struct nf_conntrack_helper *helper;
  67         struct nf_conntrack_zone zone;
  68         struct nf_conn *ct;
  69         u8 commit : 1;
  70         u8 nat : 3;                 /* enum ovs_ct_nat */
  71         u8 force : 1;
  72         u8 have_eventmask : 1;
  73         u16 family;
  74         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  75         struct md_mark mark;
  76         struct md_labels labels;
  77 #ifdef CONFIG_NF_NAT_NEEDED
  78         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  79 #endif
  80 };
  81
  82 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  83 #define OVS_CT_LIMIT_UNLIMITED  0
  84 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  85 #define CT_LIMIT_HASH_BUCKETS 512
  86 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  87
  88 struct ovs_ct_limit {
  89         /* Elements in ovs_ct_limit_info->limits hash table */
  90         struct hlist_node hlist_node;
  91         struct rcu_head rcu;
  92         u16 zone;
  93         u32 limit;
  94 };
  95
  96 struct ovs_ct_limit_info {
  97         u32 default_limit;
  98         struct hlist_head *limits;
  99         struct nf_conncount_data *data;
 100 };
 101
 102 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
 103         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
 104 };
 105 #endif
 106
 107 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 108
 109 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 110
 111 static u16 key_to_nfproto(const struct sw_flow_key *key)
 112 {
 113         switch (ntohs(key->eth.type)) {
 114         case ETH_P_IP:
 115                 return NFPROTO_IPV4;
 116         case ETH_P_IPV6:
 117                 return NFPROTO_IPV6;
 118         default:
 119                 return NFPROTO_UNSPEC;
 120         }
 121 }
 122
 123 /* Map SKB connection state into the values used by flow definition. */
 124 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 125 {
 126         u8 ct_state = OVS_CS_F_TRACKED;
 127
 128         switch (ctinfo) {
 129         case IP_CT_ESTABLISHED_REPLY:
 130         case IP_CT_RELATED_REPLY:
 131                 ct_state |= OVS_CS_F_REPLY_DIR;
 132                 break;
 133         default:
 134                 break;
 135         }
 136
 137         switch (ctinfo) {
 138         case IP_CT_ESTABLISHED:
 139         case IP_CT_ESTABLISHED_REPLY:
 140                 ct_state |= OVS_CS_F_ESTABLISHED;
 141                 break;
 142         case IP_CT_RELATED:
 143         case IP_CT_RELATED_REPLY:
 144                 ct_state |= OVS_CS_F_RELATED;
 145                 break;
 146         case IP_CT_NEW:
 147                 ct_state |= OVS_CS_F_NEW;
 148                 break;
 149         default:
 150                 break;
 151         }
 152
 153         return ct_state;
 154 }
 155
 156 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 157 {
 158 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 159         return ct ? ct->mark : 0;
 160 #else
 161         return 0;
 162 #endif
 163 }
 164
 165 /* Guard against conntrack labels max size shrinking below 128 bits. */
 166 #if NF_CT_LABELS_MAX_SIZE < 16
 167 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 168 #endif
 169
 170 static void ovs_ct_get_labels(const struct nf_conn *ct,
 171                               struct ovs_key_ct_labels *labels)
 172 {
 173         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 174
 175         if (cl)
 176                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 177         else
 178                 memset(labels, 0, OVS_CT_LABELS_LEN);
 179 }
 180
 181 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 182                                         const struct nf_conntrack_tuple *orig,
 183                                         u8 icmp_proto)
 184 {
 185         key->ct_orig_proto = orig->dst.protonum;
 186         if (orig->dst.protonum == icmp_proto) {
 187                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 188                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 189         } else {
 190                 key->ct.orig_tp.src = orig->src.u.all;
 191                 key->ct.orig_tp.dst = orig->dst.u.all;
 192         }
 193 }
 194
 195 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 196                                 const struct nf_conntrack_zone *zone,
 197                                 const struct nf_conn *ct)
 198 {
 199         key->ct_state = state;
 200         key->ct_zone = zone->id;
 201         key->ct.mark = ovs_ct_get_mark(ct);
 202         ovs_ct_get_labels(ct, &key->ct.labels);
 203
 204         if (ct) {
 205                 const struct nf_conntrack_tuple *orig;
 206
 207                 /* Use the master if we have one. */
 208                 if (ct->master)
 209                         ct = ct->master;
 210                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 211
 212                 /* IP version must match with the master connection. */
 213                 if (key->eth.type == htons(ETH_P_IP) &&
 214                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 215                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 216                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 217                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 218                         return;
 219                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 220                            !sw_flow_key_is_nd(key) &&
 221                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 222                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 223                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 224                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 225                         return;
 226                 }
 227         }
 228         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 229          * original direction key fields.
 230          */
 231         key->ct_orig_proto = 0;
 232 }
 233
 234 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 235  * previously sent the packet to conntrack via the ct action.  If
 236  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 237  * initialized from the connection status.
 238  */
 239 static void ovs_ct_update_key(const struct sk_buff *skb,
 240                               const struct ovs_conntrack_info *info,
 241                               struct sw_flow_key *key, bool post_ct,
 242                               bool keep_nat_flags)
 243 {
 244         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 245         enum ip_conntrack_info ctinfo;
 246         struct nf_conn *ct;
 247         u8 state = 0;
 248
 249         ct = nf_ct_get(skb, &ctinfo);
 250         if (ct) {
 251                 state = ovs_ct_get_state(ctinfo);
 252                 /* All unconfirmed entries are NEW connections. */
 253                 if (!nf_ct_is_confirmed(ct))
 254                         state |= OVS_CS_F_NEW;
 255                 /* OVS persists the related flag for the duration of the
 256                  * connection.
 257                  */
 258                 if (ct->master)
 259                         state |= OVS_CS_F_RELATED;
 260                 if (keep_nat_flags) {
 261                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 262                 } else {
 263                         if (ct->status & IPS_SRC_NAT)
 264                                 state |= OVS_CS_F_SRC_NAT;
 265                         if (ct->status & IPS_DST_NAT)
 266                                 state |= OVS_CS_F_DST_NAT;
 267                 }
 268                 zone = nf_ct_zone(ct);
 269         } else if (post_ct) {
 270                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 271                 if (info)
 272                         zone = &info->zone;
 273         }
 274         __ovs_ct_update_key(key, state, zone, ct);
 275 }
 276
 277 /* This is called to initialize CT key fields possibly coming in from the local
 278  * stack.
 279  */
 280 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 281 {
 282         ovs_ct_update_key(skb, NULL, key, false, false);
 283 }
 284
 285 #define IN6_ADDR_INITIALIZER(ADDR) \
 286         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 287           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 288
 289 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 290                    const struct sw_flow_key *output, struct sk_buff *skb)
 291 {
 292         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 293                 return -EMSGSIZE;
 294
 295         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 296             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 297                 return -EMSGSIZE;
 298
 299         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 300             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 301                 return -EMSGSIZE;
 302
 303         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 304             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 305                     &output->ct.labels))
 306                 return -EMSGSIZE;
 307
 308         if (swkey->ct_orig_proto) {
 309                 if (swkey->eth.type == htons(ETH_P_IP)) {
 310                         struct ovs_key_ct_tuple_ipv4 orig = {
 311                                 output->ipv4.ct_orig.src,
 312                                 output->ipv4.ct_orig.dst,
 313                                 output->ct.orig_tp.src,
 314                                 output->ct.orig_tp.dst,
 315                                 output->ct_orig_proto,
 316                         };
 317                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 318                                     sizeof(orig), &orig))
 319                                 return -EMSGSIZE;
 320                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 321                         struct ovs_key_ct_tuple_ipv6 orig = {
 322                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 323                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 324                                 output->ct.orig_tp.src,
 325                                 output->ct.orig_tp.dst,
 326                                 output->ct_orig_proto,
 327                         };
 328                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 329                                     sizeof(orig), &orig))
 330                                 return -EMSGSIZE;
 331                 }
 332         }
 333
 334         return 0;
 335 }
 336
 337 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 338                            u32 ct_mark, u32 mask)
 339 {
 340 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 341         u32 new_mark;
 342
 343         new_mark = ct_mark | (ct->mark & ~(mask));
 344         if (ct->mark != new_mark) {
 345                 ct->mark = new_mark;
 346                 if (nf_ct_is_confirmed(ct))
 347                         nf_conntrack_event_cache(IPCT_MARK, ct);
 348                 key->ct.mark = new_mark;
 349         }
 350
 351         return 0;
 352 #else
 353         return -ENOTSUPP;
 354 #endif
 355 }
 356
 357 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 358 {
 359         struct nf_conn_labels *cl;
 360
 361         cl = nf_ct_labels_find(ct);
 362         if (!cl) {
 363                 nf_ct_labels_ext_add(ct);
 364                 cl = nf_ct_labels_find(ct);
 365         }
 366
 367         return cl;
 368 }
 369
 370 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 371  * since the new connection is not yet confirmed, and thus no-one else has
 372  * access to it's labels, we simply write them over.
 373  */
 374 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 375                               const struct ovs_key_ct_labels *labels,
 376                               const struct ovs_key_ct_labels *mask)
 377 {
 378         struct nf_conn_labels *cl, *master_cl;
 379         bool have_mask = labels_nonzero(mask);
 380
 381         /* Inherit master's labels to the related connection? */
 382         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 383
 384         if (!master_cl && !have_mask)
 385                 return 0;   /* Nothing to do. */
 386
 387         cl = ovs_ct_get_conn_labels(ct);
 388         if (!cl)
 389                 return -ENOSPC;
 390
 391         /* Inherit the master's labels, if any. */
 392         if (master_cl)
 393                 *cl = *master_cl;
 394
 395         if (have_mask) {
 396                 u32 *dst = (u32 *)cl->bits;
 397                 int i;
 398
 399                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 400                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 401                                 (labels->ct_labels_32[i]
 402                                  & mask->ct_labels_32[i]);
 403         }
 404
 405         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 406          * IPCT_LABEL bit is set in the event cache.
 407          */
 408         nf_conntrack_event_cache(IPCT_LABEL, ct);
 409
 410         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 411
 412         return 0;
 413 }
 414
 415 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 416                              const struct ovs_key_ct_labels *labels,
 417                              const struct ovs_key_ct_labels *mask)
 418 {
 419         struct nf_conn_labels *cl;
 420         int err;
 421
 422         cl = ovs_ct_get_conn_labels(ct);
 423         if (!cl)
 424                 return -ENOSPC;
 425
 426         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 427                                     mask->ct_labels_32,
 428                                     OVS_CT_LABELS_LEN_32);
 429         if (err)
 430                 return err;
 431
 432         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 433
 434         return 0;
 435 }
 436
 437 /* 'skb' should already be pulled to nh_ofs. */
 438 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 439 {
 440         const struct nf_conntrack_helper *helper;
 441         const struct nf_conn_help *help;
 442         enum ip_conntrack_info ctinfo;
 443         unsigned int protoff;
 444         struct nf_conn *ct;
 445         int err;
 446
 447         ct = nf_ct_get(skb, &ctinfo);
 448         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 449                 return NF_ACCEPT;
 450
 451         help = nfct_help(ct);
 452         if (!help)
 453                 return NF_ACCEPT;
 454
 455         helper = rcu_dereference(help->helper);
 456         if (!helper)
 457                 return NF_ACCEPT;
 458
 459         switch (proto) {
 460         case NFPROTO_IPV4:
 461                 protoff = ip_hdrlen(skb);
 462                 break;
 463         case NFPROTO_IPV6: {
 464                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 465                 __be16 frag_off;
 466                 int ofs;
 467
 468                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 469                                        &frag_off);
 470                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 471                         pr_debug("proto header not found\n");
 472                         return NF_ACCEPT;
 473                 }
 474                 protoff = ofs;
 475                 break;
 476         }
 477         default:
 478                 WARN_ONCE(1, "helper invoked on non-IP family!");
 479                 return NF_DROP;
 480         }
 481
 482         err = helper->help(skb, protoff, ct, ctinfo);
 483         if (err != NF_ACCEPT)
 484                 return err;
 485
 486         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 487          * FTP with NAT) adusting the TCP payload size when mangling IP
 488          * addresses and/or port numbers in the text-based control connection.
 489          */
 490         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 491             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 492                 return NF_DROP;
 493         return NF_ACCEPT;
 494 }
 495
 496 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 497  * value if 'skb' is freed.
 498  */
 499 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 500                             u16 zone, struct sk_buff *skb)
 501 {
 502         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 503         int err;
 504
 505         if (key->eth.type == htons(ETH_P_IP)) {
 506                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 507
 508                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 509                 err = ip_defrag(net, skb, user);
 510                 if (err)
 511                         return err;
 512
 513                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 514 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 515         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 516                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 517
 518                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 519                 err = nf_ct_frag6_gather(net, skb, user);
 520                 if (err) {
 521                         if (err != -EINPROGRESS)
 522                                 kfree_skb(skb);
 523                         return err;
 524                 }
 525
 526                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 527                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 528 #endif
 529         } else {
 530                 kfree_skb(skb);
 531                 return -EPFNOSUPPORT;
 532         }
 533
 534         key->ip.frag = OVS_FRAG_TYPE_NONE;
 535         skb_clear_hash(skb);
 536         skb->ignore_df = 1;
 537         *OVS_CB(skb) = ovs_cb;
 538
 539         return 0;
 540 }
 541
 542 static struct nf_conntrack_expect *
 543 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 544                    u16 proto, const struct sk_buff *skb)
 545 {
 546         struct nf_conntrack_tuple tuple;
 547         struct nf_conntrack_expect *exp;
 548
 549         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 550                 return NULL;
 551
 552         exp = __nf_ct_expect_find(net, zone, &tuple);
 553         if (exp) {
 554                 struct nf_conntrack_tuple_hash *h;
 555
 556                 /* Delete existing conntrack entry, if it clashes with the
 557                  * expectation.  This can happen since conntrack ALGs do not
 558                  * check for clashes between (new) expectations and existing
 559                  * conntrack entries.  nf_conntrack_in() will check the
 560                  * expectations only if a conntrack entry can not be found,
 561                  * which can lead to OVS finding the expectation (here) in the
 562                  * init direction, but which will not be removed by the
 563                  * nf_conntrack_in() call, if a matching conntrack entry is
 564                  * found instead.  In this case all init direction packets
 565                  * would be reported as new related packets, while reply
 566                  * direction packets would be reported as un-related
 567                  * established packets.
 568                  */
 569                 h = nf_conntrack_find_get(net, zone, &tuple);
 570                 if (h) {
 571                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 572
 573                         nf_ct_delete(ct, 0, 0);
 574                         nf_conntrack_put(&ct->ct_general);
 575                 }
 576         }
 577
 578         return exp;
 579 }
 580
 581 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 582 static enum ip_conntrack_info
 583 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 584 {
 585         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 586
 587         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 588                 return IP_CT_ESTABLISHED_REPLY;
 589         /* Once we've had two way comms, always ESTABLISHED. */
 590         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 591                 return IP_CT_ESTABLISHED;
 592         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 593                 return IP_CT_RELATED;
 594         return IP_CT_NEW;
 595 }
 596
 597 /* Find an existing connection which this packet belongs to without
 598  * re-attributing statistics or modifying the connection state.  This allows an
 599  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 600  *
 601  * Must be called with rcu_read_lock.
 602  *
 603  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 604  * if there is no existing entry.
 605  */
 606 static struct nf_conn *
 607 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 608                      u8 l3num, struct sk_buff *skb, bool natted)
 609 {
 610         const struct nf_conntrack_l3proto *l3proto;
 611         const struct nf_conntrack_l4proto *l4proto;
 612         struct nf_conntrack_tuple tuple;
 613         struct nf_conntrack_tuple_hash *h;
 614         struct nf_conn *ct;
 615         unsigned int dataoff;
 616         u8 protonum;
 617
 618         l3proto = __nf_ct_l3proto_find(l3num);
 619         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 620                                  &protonum) <= 0) {
 621                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 622                 return NULL;
 623         }
 624         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 625         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 626                              protonum, net, &tuple, l3proto, l4proto)) {
 627                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 628                 return NULL;
 629         }
 630
 631         /* Must invert the tuple if skb has been transformed by NAT. */
 632         if (natted) {
 633                 struct nf_conntrack_tuple inverse;
 634
 635                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 636                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 637                         return NULL;
 638                 }
 639                 tuple = inverse;
 640         }
 641
 642         /* look for tuple match */
 643         h = nf_conntrack_find_get(net, zone, &tuple);
 644         if (!h)
 645                 return NULL;   /* Not found. */
 646
 647         ct = nf_ct_tuplehash_to_ctrack(h);
 648
 649         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 650          * select the other tuplehash to get the right 'ctinfo' bits for this
 651          * packet.
 652          */
 653         if (natted)
 654                 h = &ct->tuplehash[!h->tuple.dst.dir];
 655
 656         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 657         return ct;
 658 }
 659
 660 static
 661 struct nf_conn *ovs_ct_executed(struct net *net,
 662                                 const struct sw_flow_key *key,
 663                                 const struct ovs_conntrack_info *info,
 664                                 struct sk_buff *skb,
 665                                 bool *ct_executed)
 666 {
 667         struct nf_conn *ct = NULL;
 668
 669         /* If no ct, check if we have evidence that an existing conntrack entry
 670          * might be found for this skb.  This happens when we lose a skb->_nfct
 671          * due to an upcall, or if the direction is being forced.  If the
 672          * connection was not confirmed, it is not cached and needs to be run
 673          * through conntrack again.
 674          */
 675         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 676                        !(key->ct_state & OVS_CS_F_INVALID) &&
 677                        (key->ct_zone == info->zone.id);
 678
 679         if (*ct_executed || (!key->ct_state && info->force)) {
 680                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 681                                           !!(key->ct_state &
 682                                           OVS_CS_F_NAT_MASK));
 683         }
 684
 685         return ct;
 686 }
 687
 688 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 689 static bool skb_nfct_cached(struct net *net,
 690                             const struct sw_flow_key *key,
 691                             const struct ovs_conntrack_info *info,
 692                             struct sk_buff *skb)
 693 {
 694         enum ip_conntrack_info ctinfo;
 695         struct nf_conn *ct;
 696         bool ct_executed = true;
 697
 698         ct = nf_ct_get(skb, &ctinfo);
 699         if (!ct)
 700                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 701
 702         if (ct)
 703                 nf_ct_get(skb, &ctinfo);
 704         else
 705                 return false;
 706
 707         if (!net_eq(net, read_pnet(&ct->ct_net)))
 708                 return false;
 709         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 710                 return false;
 711         if (info->helper) {
 712                 struct nf_conn_help *help;
 713
 714                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 715                 if (help && rcu_access_pointer(help->helper) != info->helper)
 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_conntrack_put(&ct->ct_general);
 727                 nf_ct_set(skb, NULL, 0);
 728                 return false;
 729         }
 730
 731         return ct_executed;
 732 }
 733
 734 #ifdef CONFIG_NF_NAT_NEEDED
 735 /* Modelled after nf_nat_ipv[46]_fn().
 736  * range is only used for new, uninitialized NAT state.
 737  * Returns either NF_ACCEPT or NF_DROP.
 738  */
 739 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 740                               enum ip_conntrack_info ctinfo,
 741                               const struct nf_nat_range2 *range,
 742                               enum nf_nat_manip_type maniptype)
 743 {
 744         int hooknum, nh_off, err = NF_ACCEPT;
 745
 746         nh_off = skb_network_offset(skb);
 747         skb_pull_rcsum(skb, nh_off);
 748
 749         /* See HOOK2MANIP(). */
 750         if (maniptype == NF_NAT_MANIP_SRC)
 751                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 752         else
 753                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 754
 755         switch (ctinfo) {
 756         case IP_CT_RELATED:
 757         case IP_CT_RELATED_REPLY:
 758                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 759                     skb->protocol == htons(ETH_P_IP) &&
 760                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 761                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 762                                                            hooknum))
 763                                 err = NF_DROP;
 764                         goto push;
 765                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 766                            skb->protocol == htons(ETH_P_IPV6)) {
 767                         __be16 frag_off;
 768                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 769                         int hdrlen = ipv6_skip_exthdr(skb,
 770                                                       sizeof(struct ipv6hdr),
 771                                                       &nexthdr, &frag_off);
 772
 773                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 774                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 775                                                                      ctinfo,
 776                                                                      hooknum,
 777                                                                      hdrlen))
 778                                         err = NF_DROP;
 779                                 goto push;
 780                         }
 781                 }
 782                 /* Non-ICMP, fall thru to initialize if needed. */
 783                 /* fall through */
 784         case IP_CT_NEW:
 785                 /* Seen it before?  This can happen for loopback, retrans,
 786                  * or local packets.
 787                  */
 788                 if (!nf_nat_initialized(ct, maniptype)) {
 789                         /* Initialize according to the NAT action. */
 790                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 791                                 /* Action is set up to establish a new
 792                                  * mapping.
 793                                  */
 794                                 ? nf_nat_setup_info(ct, range, maniptype)
 795                                 : nf_nat_alloc_null_binding(ct, hooknum);
 796                         if (err != NF_ACCEPT)
 797                                 goto push;
 798                 }
 799                 break;
 800
 801         case IP_CT_ESTABLISHED:
 802         case IP_CT_ESTABLISHED_REPLY:
 803                 break;
 804
 805         default:
 806                 err = NF_DROP;
 807                 goto push;
 808         }
 809
 810         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 811 push:
 812         skb_push(skb, nh_off);
 813         skb_postpush_rcsum(skb, skb->data, nh_off);
 814
 815         return err;
 816 }
 817
 818 static void ovs_nat_update_key(struct sw_flow_key *key,
 819                                const struct sk_buff *skb,
 820                                enum nf_nat_manip_type maniptype)
 821 {
 822         if (maniptype == NF_NAT_MANIP_SRC) {
 823                 __be16 src;
 824
 825                 key->ct_state |= OVS_CS_F_SRC_NAT;
 826                 if (key->eth.type == htons(ETH_P_IP))
 827                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 828                 else if (key->eth.type == htons(ETH_P_IPV6))
 829                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 830                                sizeof(key->ipv6.addr.src));
 831                 else
 832                         return;
 833
 834                 if (key->ip.proto == IPPROTO_UDP)
 835                         src = udp_hdr(skb)->source;
 836                 else if (key->ip.proto == IPPROTO_TCP)
 837                         src = tcp_hdr(skb)->source;
 838                 else if (key->ip.proto == IPPROTO_SCTP)
 839                         src = sctp_hdr(skb)->source;
 840                 else
 841                         return;
 842
 843                 key->tp.src = src;
 844         } else {
 845                 __be16 dst;
 846
 847                 key->ct_state |= OVS_CS_F_DST_NAT;
 848                 if (key->eth.type == htons(ETH_P_IP))
 849                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 850                 else if (key->eth.type == htons(ETH_P_IPV6))
 851                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 852                                sizeof(key->ipv6.addr.dst));
 853                 else
 854                         return;
 855
 856                 if (key->ip.proto == IPPROTO_UDP)
 857                         dst = udp_hdr(skb)->dest;
 858                 else if (key->ip.proto == IPPROTO_TCP)
 859                         dst = tcp_hdr(skb)->dest;
 860                 else if (key->ip.proto == IPPROTO_SCTP)
 861                         dst = sctp_hdr(skb)->dest;
 862                 else
 863                         return;
 864
 865                 key->tp.dst = dst;
 866         }
 867 }
 868
 869 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 870 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 871                       const struct ovs_conntrack_info *info,
 872                       struct sk_buff *skb, struct nf_conn *ct,
 873                       enum ip_conntrack_info ctinfo)
 874 {
 875         enum nf_nat_manip_type maniptype;
 876         int err;
 877
 878         /* Add NAT extension if not confirmed yet. */
 879         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 880                 return NF_ACCEPT;   /* Can't NAT. */
 881
 882         /* Determine NAT type.
 883          * Check if the NAT type can be deduced from the tracked connection.
 884          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 885          * when committing.
 886          */
 887         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 888             ct->status & IPS_NAT_MASK &&
 889             (ctinfo != IP_CT_RELATED || info->commit)) {
 890                 /* NAT an established or related connection like before. */
 891                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 892                         /* This is the REPLY direction for a connection
 893                          * for which NAT was applied in the forward
 894                          * direction.  Do the reverse NAT.
 895                          */
 896                         maniptype = ct->status & IPS_SRC_NAT
 897                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 898                 else
 899                         maniptype = ct->status & IPS_SRC_NAT
 900                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 901         } else if (info->nat & OVS_CT_SRC_NAT) {
 902                 maniptype = NF_NAT_MANIP_SRC;
 903         } else if (info->nat & OVS_CT_DST_NAT) {
 904                 maniptype = NF_NAT_MANIP_DST;
 905         } else {
 906                 return NF_ACCEPT; /* Connection is not NATed. */
 907         }
 908         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 909
 910         /* Mark NAT done if successful and update the flow key. */
 911         if (err == NF_ACCEPT)
 912                 ovs_nat_update_key(key, skb, maniptype);
 913
 914         return err;
 915 }
 916 #else /* !CONFIG_NF_NAT_NEEDED */
 917 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 918                       const struct ovs_conntrack_info *info,
 919                       struct sk_buff *skb, struct nf_conn *ct,
 920                       enum ip_conntrack_info ctinfo)
 921 {
 922         return NF_ACCEPT;
 923 }
 924 #endif
 925
 926 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 927  * not done already.  Update key with new CT state after passing the packet
 928  * through conntrack.
 929  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 930  * set to NULL and 0 will be returned.
 931  */
 932 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 933                            const struct ovs_conntrack_info *info,
 934                            struct sk_buff *skb)
 935 {
 936         /* If we are recirculating packets to match on conntrack fields and
 937          * committing with a separate conntrack action,  then we don't need to
 938          * actually run the packet through conntrack twice unless it's for a
 939          * different zone.
 940          */
 941         bool cached = skb_nfct_cached(net, key, info, skb);
 942         enum ip_conntrack_info ctinfo;
 943         struct nf_conn *ct;
 944
 945         if (!cached) {
 946                 struct nf_conn *tmpl = info->ct;
 947                 int err;
 948
 949                 /* Associate skb with specified zone. */
 950                 if (tmpl) {
 951                         if (skb_nfct(skb))
 952                                 nf_conntrack_put(skb_nfct(skb));
 953                         nf_conntrack_get(&tmpl->ct_general);
 954                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 955                 }
 956
 957                 err = nf_conntrack_in(net, info->family,
 958                                       NF_INET_PRE_ROUTING, skb);
 959                 if (err != NF_ACCEPT)
 960                         return -ENOENT;
 961
 962                 /* Clear CT state NAT flags to mark that we have not yet done
 963                  * NAT after the nf_conntrack_in() call.  We can actually clear
 964                  * the whole state, as it will be re-initialized below.
 965                  */
 966                 key->ct_state = 0;
 967
 968                 /* Update the key, but keep the NAT flags. */
 969                 ovs_ct_update_key(skb, info, key, true, true);
 970         }
 971
 972         ct = nf_ct_get(skb, &ctinfo);
 973         if (ct) {
 974                 /* Packets starting a new connection must be NATted before the
 975                  * helper, so that the helper knows about the NAT.  We enforce
 976                  * this by delaying both NAT and helper calls for unconfirmed
 977                  * connections until the committing CT action.  For later
 978                  * packets NAT and Helper may be called in either order.
 979                  *
 980                  * NAT will be done only if the CT action has NAT, and only
 981                  * once per packet (per zone), as guarded by the NAT bits in
 982                  * the key->ct_state.
 983                  */
 984                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 985                     (nf_ct_is_confirmed(ct) || info->commit) &&
 986                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 987                         return -EINVAL;
 988                 }
 989
 990                 /* Userspace may decide to perform a ct lookup without a helper
 991                  * specified followed by a (recirculate and) commit with one.
 992                  * Therefore, for unconfirmed connections which we will commit,
 993                  * we need to attach the helper here.
 994                  */
 995                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 996                     info->helper && !nfct_help(ct)) {
 997                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 998                                                             GFP_ATOMIC);
 999                         if (err)
1000                                 return err;
1001                 }
1002
1003                 /* Call the helper only if:
1004                  * - nf_conntrack_in() was executed above ("!cached") for a
1005                  *   confirmed connection, or
1006                  * - When committing an unconfirmed connection.
1007                  */
1008                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1009                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1010                         return -EINVAL;
1011                 }
1012         }
1013
1014         return 0;
1015 }
1016
1017 /* Lookup connection and read fields into key. */
1018 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1019                          const struct ovs_conntrack_info *info,
1020                          struct sk_buff *skb)
1021 {
1022         struct nf_conntrack_expect *exp;
1023
1024         /* If we pass an expected packet through nf_conntrack_in() the
1025          * expectation is typically removed, but the packet could still be
1026          * lost in upcall processing.  To prevent this from happening we
1027          * perform an explicit expectation lookup.  Expected connections are
1028          * always new, and will be passed through conntrack only when they are
1029          * committed, as it is OK to remove the expectation at that time.
1030          */
1031         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1032         if (exp) {
1033                 u8 state;
1034
1035                 /* NOTE: New connections are NATted and Helped only when
1036                  * committed, so we are not calling into NAT here.
1037                  */
1038                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1039                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1040         } else {
1041                 struct nf_conn *ct;
1042                 int err;
1043
1044                 err = __ovs_ct_lookup(net, key, info, skb);
1045                 if (err)
1046                         return err;
1047
1048                 ct = (struct nf_conn *)skb_nfct(skb);
1049                 if (ct)
1050                         nf_ct_deliver_cached_events(ct);
1051         }
1052
1053         return 0;
1054 }
1055
1056 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1057 {
1058         size_t i;
1059
1060         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1061                 if (labels->ct_labels_32[i])
1062                         return true;
1063
1064         return false;
1065 }
1066
1067 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1068 static struct hlist_head *ct_limit_hash_bucket(
1069         const struct ovs_ct_limit_info *info, u16 zone)
1070 {
1071         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1072 }
1073
1074 /* Call with ovs_mutex */
1075 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1076                          struct ovs_ct_limit *new_ct_limit)
1077 {
1078         struct ovs_ct_limit *ct_limit;
1079         struct hlist_head *head;
1080
1081         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1082         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1083                 if (ct_limit->zone == new_ct_limit->zone) {
1084                         hlist_replace_rcu(&ct_limit->hlist_node,
1085                                           &new_ct_limit->hlist_node);
1086                         kfree_rcu(ct_limit, rcu);
1087                         return;
1088                 }
1089         }
1090
1091         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1092 }
1093
1094 /* Call with ovs_mutex */
1095 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1096 {
1097         struct ovs_ct_limit *ct_limit;
1098         struct hlist_head *head;
1099         struct hlist_node *n;
1100
1101         head = ct_limit_hash_bucket(info, zone);
1102         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1103                 if (ct_limit->zone == zone) {
1104                         hlist_del_rcu(&ct_limit->hlist_node);
1105                         kfree_rcu(ct_limit, rcu);
1106                         return;
1107                 }
1108         }
1109 }
1110
1111 /* Call with RCU read lock */
1112 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1113 {
1114         struct ovs_ct_limit *ct_limit;
1115         struct hlist_head *head;
1116
1117         head = ct_limit_hash_bucket(info, zone);
1118         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1119                 if (ct_limit->zone == zone)
1120                         return ct_limit->limit;
1121         }
1122
1123         return info->default_limit;
1124 }
1125
1126 static int ovs_ct_check_limit(struct net *net,
1127                               const struct ovs_conntrack_info *info,
1128                               const struct nf_conntrack_tuple *tuple)
1129 {
1130         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1131         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1132         u32 per_zone_limit, connections;
1133         u32 conncount_key;
1134
1135         conncount_key = info->zone.id;
1136
1137         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1138         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1139                 return 0;
1140
1141         connections = nf_conncount_count(net, ct_limit_info->data,
1142                                          &conncount_key, tuple, &info->zone);
1143         if (connections > per_zone_limit)
1144                 return -ENOMEM;
1145
1146         return 0;
1147 }
1148 #endif
1149
1150 /* Lookup connection and confirm if unconfirmed. */
1151 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1152                          const struct ovs_conntrack_info *info,
1153                          struct sk_buff *skb)
1154 {
1155         enum ip_conntrack_info ctinfo;
1156         struct nf_conn *ct;
1157         int err;
1158
1159         err = __ovs_ct_lookup(net, key, info, skb);
1160         if (err)
1161                 return err;
1162
1163         /* The connection could be invalid, in which case this is a no-op.*/
1164         ct = nf_ct_get(skb, &ctinfo);
1165         if (!ct)
1166                 return 0;
1167
1168 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1169         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1170                 if (!nf_ct_is_confirmed(ct)) {
1171                         err = ovs_ct_check_limit(net, info,
1172                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1173                         if (err) {
1174                                 net_warn_ratelimited("openvswitch: zone: %u "
1175                                         "execeeds conntrack limit\n",
1176                                         info->zone.id);
1177                                 return err;
1178                         }
1179                 }
1180         }
1181 #endif
1182
1183         /* Set the conntrack event mask if given.  NEW and DELETE events have
1184          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1185          * typically would receive many kinds of updates.  Setting the event
1186          * mask allows those events to be filtered.  The set event mask will
1187          * remain in effect for the lifetime of the connection unless changed
1188          * by a further CT action with both the commit flag and the eventmask
1189          * option. */
1190         if (info->have_eventmask) {
1191                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1192
1193                 if (cache)
1194                         cache->ctmask = info->eventmask;
1195         }
1196
1197         /* Apply changes before confirming the connection so that the initial
1198          * conntrack NEW netlink event carries the values given in the CT
1199          * action.
1200          */
1201         if (info->mark.mask) {
1202                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1203                                       info->mark.mask);
1204                 if (err)
1205                         return err;
1206         }
1207         if (!nf_ct_is_confirmed(ct)) {
1208                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1209                                          &info->labels.mask);
1210                 if (err)
1211                         return err;
1212         } else if (labels_nonzero(&info->labels.mask)) {
1213                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1214                                         &info->labels.mask);
1215                 if (err)
1216                         return err;
1217         }
1218         /* This will take care of sending queued events even if the connection
1219          * is already confirmed.
1220          */
1221         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1222                 return -EINVAL;
1223
1224         return 0;
1225 }
1226
1227 /* Trim the skb to the length specified by the IP/IPv6 header,
1228  * removing any trailing lower-layer padding. This prepares the skb
1229  * for higher-layer processing that assumes skb->len excludes padding
1230  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1231  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1232  */
1233 static int ovs_skb_network_trim(struct sk_buff *skb)
1234 {
1235         unsigned int len;
1236         int err;
1237
1238         switch (skb->protocol) {
1239         case htons(ETH_P_IP):
1240                 len = ntohs(ip_hdr(skb)->tot_len);
1241                 break;
1242         case htons(ETH_P_IPV6):
1243                 len = sizeof(struct ipv6hdr)
1244                         + ntohs(ipv6_hdr(skb)->payload_len);
1245                 break;
1246         default:
1247                 len = skb->len;
1248         }
1249
1250         err = pskb_trim_rcsum(skb, len);
1251         if (err)
1252                 kfree_skb(skb);
1253
1254         return err;
1255 }
1256
1257 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1258  * value if 'skb' is freed.
1259  */
1260 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1261                    struct sw_flow_key *key,
1262                    const struct ovs_conntrack_info *info)
1263 {
1264         int nh_ofs;
1265         int err;
1266
1267         /* The conntrack module expects to be working at L3. */
1268         nh_ofs = skb_network_offset(skb);
1269         skb_pull_rcsum(skb, nh_ofs);
1270
1271         err = ovs_skb_network_trim(skb);
1272         if (err)
1273                 return err;
1274
1275         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1276                 err = handle_fragments(net, key, info->zone.id, skb);
1277                 if (err)
1278                         return err;
1279         }
1280
1281         if (info->commit)
1282                 err = ovs_ct_commit(net, key, info, skb);
1283         else
1284                 err = ovs_ct_lookup(net, key, info, skb);
1285
1286         skb_push(skb, nh_ofs);
1287         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1288         if (err)
1289                 kfree_skb(skb);
1290         return err;
1291 }
1292
1293 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1294 {
1295         if (skb_nfct(skb)) {
1296                 nf_conntrack_put(skb_nfct(skb));
1297                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1298                 ovs_ct_fill_key(skb, key);
1299         }
1300
1301         return 0;
1302 }
1303
1304 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1305                              const struct sw_flow_key *key, bool log)
1306 {
1307         struct nf_conntrack_helper *helper;
1308         struct nf_conn_help *help;
1309
1310         helper = nf_conntrack_helper_try_module_get(name, info->family,
1311                                                     key->ip.proto);
1312         if (!helper) {
1313                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1314                 return -EINVAL;
1315         }
1316
1317         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1318         if (!help) {
1319                 nf_conntrack_helper_put(helper);
1320                 return -ENOMEM;
1321         }
1322
1323         rcu_assign_pointer(help->helper, helper);
1324         info->helper = helper;
1325         return 0;
1326 }
1327
1328 #ifdef CONFIG_NF_NAT_NEEDED
1329 static int parse_nat(const struct nlattr *attr,
1330                      struct ovs_conntrack_info *info, bool log)
1331 {
1332         struct nlattr *a;
1333         int rem;
1334         bool have_ip_max = false;
1335         bool have_proto_max = false;
1336         bool ip_vers = (info->family == NFPROTO_IPV6);
1337
1338         nla_for_each_nested(a, attr, rem) {
1339                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1340                         [OVS_NAT_ATTR_SRC] = {0, 0},
1341                         [OVS_NAT_ATTR_DST] = {0, 0},
1342                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1343                                                  sizeof(struct in6_addr)},
1344                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1345                                                  sizeof(struct in6_addr)},
1346                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1347                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1348                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1349                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1350                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1351                 };
1352                 int type = nla_type(a);
1353
1354                 if (type > OVS_NAT_ATTR_MAX) {
1355                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1356                                   type, OVS_NAT_ATTR_MAX);
1357                         return -EINVAL;
1358                 }
1359
1360                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1361                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1362                                   type, nla_len(a),
1363                                   ovs_nat_attr_lens[type][ip_vers]);
1364                         return -EINVAL;
1365                 }
1366
1367                 switch (type) {
1368                 case OVS_NAT_ATTR_SRC:
1369                 case OVS_NAT_ATTR_DST:
1370                         if (info->nat) {
1371                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1372                                 return -ERANGE;
1373                         }
1374                         info->nat |= OVS_CT_NAT;
1375                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1376                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1377                         break;
1378
1379                 case OVS_NAT_ATTR_IP_MIN:
1380                         nla_memcpy(&info->range.min_addr, a,
1381                                    sizeof(info->range.min_addr));
1382                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1383                         break;
1384
1385                 case OVS_NAT_ATTR_IP_MAX:
1386                         have_ip_max = true;
1387                         nla_memcpy(&info->range.max_addr, a,
1388                                    sizeof(info->range.max_addr));
1389                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1390                         break;
1391
1392                 case OVS_NAT_ATTR_PROTO_MIN:
1393                         info->range.min_proto.all = htons(nla_get_u16(a));
1394                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1395                         break;
1396
1397                 case OVS_NAT_ATTR_PROTO_MAX:
1398                         have_proto_max = true;
1399                         info->range.max_proto.all = htons(nla_get_u16(a));
1400                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1401                         break;
1402
1403                 case OVS_NAT_ATTR_PERSISTENT:
1404                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1405                         break;
1406
1407                 case OVS_NAT_ATTR_PROTO_HASH:
1408                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1409                         break;
1410
1411                 case OVS_NAT_ATTR_PROTO_RANDOM:
1412                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1413                         break;
1414
1415                 default:
1416                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1417                         return -EINVAL;
1418                 }
1419         }
1420
1421         if (rem > 0) {
1422                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1423                 return -EINVAL;
1424         }
1425         if (!info->nat) {
1426                 /* Do not allow flags if no type is given. */
1427                 if (info->range.flags) {
1428                         OVS_NLERR(log,
1429                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1430                                   );
1431                         return -EINVAL;
1432                 }
1433                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1434         } else if (!info->commit) {
1435                 OVS_NLERR(log,
1436                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1437                           );
1438                 return -EINVAL;
1439         }
1440         /* Allow missing IP_MAX. */
1441         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1442                 memcpy(&info->range.max_addr, &info->range.min_addr,
1443                        sizeof(info->range.max_addr));
1444         }
1445         /* Allow missing PROTO_MAX. */
1446         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1447             !have_proto_max) {
1448                 info->range.max_proto.all = info->range.min_proto.all;
1449         }
1450         return 0;
1451 }
1452 #endif
1453
1454 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1455         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1456         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1457         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1458                                     .maxlen = sizeof(u16) },
1459         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1460                                     .maxlen = sizeof(struct md_mark) },
1461         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1462                                     .maxlen = sizeof(struct md_labels) },
1463         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1464                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1465 #ifdef CONFIG_NF_NAT_NEEDED
1466         /* NAT length is checked when parsing the nested attributes. */
1467         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1468 #endif
1469         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1470                                     .maxlen = sizeof(u32) },
1471 };
1472
1473 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1474                     const char **helper, bool log)
1475 {
1476         struct nlattr *a;
1477         int rem;
1478
1479         nla_for_each_nested(a, attr, rem) {
1480                 int type = nla_type(a);
1481                 int maxlen;
1482                 int minlen;
1483
1484                 if (type > OVS_CT_ATTR_MAX) {
1485                         OVS_NLERR(log,
1486                                   "Unknown conntrack attr (type=%d, max=%d)",
1487                                   type, OVS_CT_ATTR_MAX);
1488                         return -EINVAL;
1489                 }
1490
1491                 maxlen = ovs_ct_attr_lens[type].maxlen;
1492                 minlen = ovs_ct_attr_lens[type].minlen;
1493                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1494                         OVS_NLERR(log,
1495                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1496                                   type, nla_len(a), maxlen);
1497                         return -EINVAL;
1498                 }
1499
1500                 switch (type) {
1501                 case OVS_CT_ATTR_FORCE_COMMIT:
1502                         info->force = true;
1503                         /* fall through. */
1504                 case OVS_CT_ATTR_COMMIT:
1505                         info->commit = true;
1506                         break;
1507 #ifdef CONFIG_NF_CONNTRACK_ZONES
1508                 case OVS_CT_ATTR_ZONE:
1509                         info->zone.id = nla_get_u16(a);
1510                         break;
1511 #endif
1512 #ifdef CONFIG_NF_CONNTRACK_MARK
1513                 case OVS_CT_ATTR_MARK: {
1514                         struct md_mark *mark = nla_data(a);
1515
1516                         if (!mark->mask) {
1517                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1518                                 return -EINVAL;
1519                         }
1520                         info->mark = *mark;
1521                         break;
1522                 }
1523 #endif
1524 #ifdef CONFIG_NF_CONNTRACK_LABELS
1525                 case OVS_CT_ATTR_LABELS: {
1526                         struct md_labels *labels = nla_data(a);
1527
1528                         if (!labels_nonzero(&labels->mask)) {
1529                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1530                                 return -EINVAL;
1531                         }
1532                         info->labels = *labels;
1533                         break;
1534                 }
1535 #endif
1536                 case OVS_CT_ATTR_HELPER:
1537                         *helper = nla_data(a);
1538                         if (!memchr(*helper, '\0', nla_len(a))) {
1539                                 OVS_NLERR(log, "Invalid conntrack helper");
1540                                 return -EINVAL;
1541                         }
1542                         break;
1543 #ifdef CONFIG_NF_NAT_NEEDED
1544                 case OVS_CT_ATTR_NAT: {
1545                         int err = parse_nat(a, info, log);
1546
1547                         if (err)
1548                                 return err;
1549                         break;
1550                 }
1551 #endif
1552                 case OVS_CT_ATTR_EVENTMASK:
1553                         info->have_eventmask = true;
1554                         info->eventmask = nla_get_u32(a);
1555                         break;
1556
1557                 default:
1558                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1559                                   type);
1560                         return -EINVAL;
1561                 }
1562         }
1563
1564 #ifdef CONFIG_NF_CONNTRACK_MARK
1565         if (!info->commit && info->mark.mask) {
1566                 OVS_NLERR(log,
1567                           "Setting conntrack mark requires 'commit' flag.");
1568                 return -EINVAL;
1569         }
1570 #endif
1571 #ifdef CONFIG_NF_CONNTRACK_LABELS
1572         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1573                 OVS_NLERR(log,
1574                           "Setting conntrack labels requires 'commit' flag.");
1575                 return -EINVAL;
1576         }
1577 #endif
1578         if (rem > 0) {
1579                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1580                 return -EINVAL;
1581         }
1582
1583         return 0;
1584 }
1585
1586 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1587 {
1588         if (attr == OVS_KEY_ATTR_CT_STATE)
1589                 return true;
1590         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1591             attr == OVS_KEY_ATTR_CT_ZONE)
1592                 return true;
1593         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1594             attr == OVS_KEY_ATTR_CT_MARK)
1595                 return true;
1596         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1597             attr == OVS_KEY_ATTR_CT_LABELS) {
1598                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1599
1600                 return ovs_net->xt_label;
1601         }
1602
1603         return false;
1604 }
1605
1606 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1607                        const struct sw_flow_key *key,
1608                        struct sw_flow_actions **sfa,  bool log)
1609 {
1610         struct ovs_conntrack_info ct_info;
1611         const char *helper = NULL;
1612         u16 family;
1613         int err;
1614
1615         family = key_to_nfproto(key);
1616         if (family == NFPROTO_UNSPEC) {
1617                 OVS_NLERR(log, "ct family unspecified");
1618                 return -EINVAL;
1619         }
1620
1621         memset(&ct_info, 0, sizeof(ct_info));
1622         ct_info.family = family;
1623
1624         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1625                         NF_CT_DEFAULT_ZONE_DIR, 0);
1626
1627         err = parse_ct(attr, &ct_info, &helper, log);
1628         if (err)
1629                 return err;
1630
1631         /* Set up template for tracking connections in specific zones. */
1632         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1633         if (!ct_info.ct) {
1634                 OVS_NLERR(log, "Failed to allocate conntrack template");
1635                 return -ENOMEM;
1636         }
1637
1638         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1639         nf_conntrack_get(&ct_info.ct->ct_general);
1640
1641         if (helper) {
1642                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1643                 if (err)
1644                         goto err_free_ct;
1645         }
1646
1647         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1648                                  sizeof(ct_info), log);
1649         if (err)
1650                 goto err_free_ct;
1651
1652         return 0;
1653 err_free_ct:
1654         __ovs_ct_free_action(&ct_info);
1655         return err;
1656 }
1657
1658 #ifdef CONFIG_NF_NAT_NEEDED
1659 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1660                                struct sk_buff *skb)
1661 {
1662         struct nlattr *start;
1663
1664         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1665         if (!start)
1666                 return false;
1667
1668         if (info->nat & OVS_CT_SRC_NAT) {
1669                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1670                         return false;
1671         } else if (info->nat & OVS_CT_DST_NAT) {
1672                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1673                         return false;
1674         } else {
1675                 goto out;
1676         }
1677
1678         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1679                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1680                     info->family == NFPROTO_IPV4) {
1681                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1682                                             info->range.min_addr.ip) ||
1683                             (info->range.max_addr.ip
1684                              != info->range.min_addr.ip &&
1685                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1686                                               info->range.max_addr.ip))))
1687                                 return false;
1688                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1689                            info->family == NFPROTO_IPV6) {
1690                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1691                                              &info->range.min_addr.in6) ||
1692                             (memcmp(&info->range.max_addr.in6,
1693                                     &info->range.min_addr.in6,
1694                                     sizeof(info->range.max_addr.in6)) &&
1695                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1696                                                &info->range.max_addr.in6))))
1697                                 return false;
1698                 } else {
1699                         return false;
1700                 }
1701         }
1702         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1703             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1704                          ntohs(info->range.min_proto.all)) ||
1705              (info->range.max_proto.all != info->range.min_proto.all &&
1706               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1707                           ntohs(info->range.max_proto.all)))))
1708                 return false;
1709
1710         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1711             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1712                 return false;
1713         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1714             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1715                 return false;
1716         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1717             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1718                 return false;
1719 out:
1720         nla_nest_end(skb, start);
1721
1722         return true;
1723 }
1724 #endif
1725
1726 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1727                           struct sk_buff *skb)
1728 {
1729         struct nlattr *start;
1730
1731         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1732         if (!start)
1733                 return -EMSGSIZE;
1734
1735         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1736                                             ? OVS_CT_ATTR_FORCE_COMMIT
1737                                             : OVS_CT_ATTR_COMMIT))
1738                 return -EMSGSIZE;
1739         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1740             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1741                 return -EMSGSIZE;
1742         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1743             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1744                     &ct_info->mark))
1745                 return -EMSGSIZE;
1746         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1747             labels_nonzero(&ct_info->labels.mask) &&
1748             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1749                     &ct_info->labels))
1750                 return -EMSGSIZE;
1751         if (ct_info->helper) {
1752                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1753                                    ct_info->helper->name))
1754                         return -EMSGSIZE;
1755         }
1756         if (ct_info->have_eventmask &&
1757             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1758                 return -EMSGSIZE;
1759
1760 #ifdef CONFIG_NF_NAT_NEEDED
1761         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1762                 return -EMSGSIZE;
1763 #endif
1764         nla_nest_end(skb, start);
1765
1766         return 0;
1767 }
1768
1769 void ovs_ct_free_action(const struct nlattr *a)
1770 {
1771         struct ovs_conntrack_info *ct_info = nla_data(a);
1772
1773         __ovs_ct_free_action(ct_info);
1774 }
1775
1776 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1777 {
1778         if (ct_info->helper)
1779                 nf_conntrack_helper_put(ct_info->helper);
1780         if (ct_info->ct)
1781                 nf_ct_tmpl_free(ct_info->ct);
1782 }
1783
1784 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1785 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1786 {
1787         int i, err;
1788
1789         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1790                                          GFP_KERNEL);
1791         if (!ovs_net->ct_limit_info)
1792                 return -ENOMEM;
1793
1794         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1795         ovs_net->ct_limit_info->limits =
1796                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1797                               GFP_KERNEL);
1798         if (!ovs_net->ct_limit_info->limits) {
1799                 kfree(ovs_net->ct_limit_info);
1800                 return -ENOMEM;
1801         }
1802
1803         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1804                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1805
1806         ovs_net->ct_limit_info->data =
1807                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1808
1809         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1810                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1811                 kfree(ovs_net->ct_limit_info->limits);
1812                 kfree(ovs_net->ct_limit_info);
1813                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1814                 return err;
1815         }
1816         return 0;
1817 }
1818
1819 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1820 {
1821         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1822         int i;
1823
1824         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1825         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1826                 struct hlist_head *head = &info->limits[i];
1827                 struct ovs_ct_limit *ct_limit;
1828
1829                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1830                         kfree_rcu(ct_limit, rcu);
1831         }
1832         kfree(ovs_net->ct_limit_info->limits);
1833         kfree(ovs_net->ct_limit_info);
1834 }
1835
1836 static struct sk_buff *
1837 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1838                              struct ovs_header **ovs_reply_header)
1839 {
1840         struct ovs_header *ovs_header = info->userhdr;
1841         struct sk_buff *skb;
1842
1843         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1844         if (!skb)
1845                 return ERR_PTR(-ENOMEM);
1846
1847         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1848                                         info->snd_seq,
1849                                         &dp_ct_limit_genl_family, 0, cmd);
1850
1851         if (!*ovs_reply_header) {
1852                 nlmsg_free(skb);
1853                 return ERR_PTR(-EMSGSIZE);
1854         }
1855         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1856
1857         return skb;
1858 }
1859
1860 static bool check_zone_id(int zone_id, u16 *pzone)
1861 {
1862         if (zone_id >= 0 && zone_id <= 65535) {
1863                 *pzone = (u16)zone_id;
1864                 return true;
1865         }
1866         return false;
1867 }
1868
1869 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1870                                        struct ovs_ct_limit_info *info)
1871 {
1872         struct ovs_zone_limit *zone_limit;
1873         int rem;
1874         u16 zone;
1875
1876         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1877         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1878
1879         while (rem >= sizeof(*zone_limit)) {
1880                 if (unlikely(zone_limit->zone_id ==
1881                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1882                         ovs_lock();
1883                         info->default_limit = zone_limit->limit;
1884                         ovs_unlock();
1885                 } else if (unlikely(!check_zone_id(
1886                                 zone_limit->zone_id, &zone))) {
1887                         OVS_NLERR(true, "zone id is out of range");
1888                 } else {
1889                         struct ovs_ct_limit *ct_limit;
1890
1891                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1892                         if (!ct_limit)
1893                                 return -ENOMEM;
1894
1895                         ct_limit->zone = zone;
1896                         ct_limit->limit = zone_limit->limit;
1897
1898                         ovs_lock();
1899                         ct_limit_set(info, ct_limit);
1900                         ovs_unlock();
1901                 }
1902                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1903                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1904                                 NLA_ALIGN(sizeof(*zone_limit)));
1905         }
1906
1907         if (rem)
1908                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1909
1910         return 0;
1911 }
1912
1913 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1914                                        struct ovs_ct_limit_info *info)
1915 {
1916         struct ovs_zone_limit *zone_limit;
1917         int rem;
1918         u16 zone;
1919
1920         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1921         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1922
1923         while (rem >= sizeof(*zone_limit)) {
1924                 if (unlikely(zone_limit->zone_id ==
1925                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1926                         ovs_lock();
1927                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1928                         ovs_unlock();
1929                 } else if (unlikely(!check_zone_id(
1930                                 zone_limit->zone_id, &zone))) {
1931                         OVS_NLERR(true, "zone id is out of range");
1932                 } else {
1933                         ovs_lock();
1934                         ct_limit_del(info, zone);
1935                         ovs_unlock();
1936                 }
1937                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1938                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1939                                 NLA_ALIGN(sizeof(*zone_limit)));
1940         }
1941
1942         if (rem)
1943                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1944
1945         return 0;
1946 }
1947
1948 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1949                                           struct sk_buff *reply)
1950 {
1951         struct ovs_zone_limit zone_limit;
1952         int err;
1953
1954         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1955         zone_limit.limit = info->default_limit;
1956         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1957         if (err)
1958                 return err;
1959
1960         return 0;
1961 }
1962
1963 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1964                                          struct nf_conncount_data *data,
1965                                          u16 zone_id, u32 limit,
1966                                          struct sk_buff *reply)
1967 {
1968         struct nf_conntrack_zone ct_zone;
1969         struct ovs_zone_limit zone_limit;
1970         u32 conncount_key = zone_id;
1971
1972         zone_limit.zone_id = zone_id;
1973         zone_limit.limit = limit;
1974         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1975
1976         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1977                                               &ct_zone);
1978         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1979 }
1980
1981 static int ovs_ct_limit_get_zone_limit(struct net *net,
1982                                        struct nlattr *nla_zone_limit,
1983                                        struct ovs_ct_limit_info *info,
1984                                        struct sk_buff *reply)
1985 {
1986         struct ovs_zone_limit *zone_limit;
1987         int rem, err;
1988         u32 limit;
1989         u16 zone;
1990
1991         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1992         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1993
1994         while (rem >= sizeof(*zone_limit)) {
1995                 if (unlikely(zone_limit->zone_id ==
1996                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1997                         err = ovs_ct_limit_get_default_limit(info, reply);
1998                         if (err)
1999                                 return err;
2000                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2001                                                         &zone))) {
2002                         OVS_NLERR(true, "zone id is out of range");
2003                 } else {
2004                         rcu_read_lock();
2005                         limit = ct_limit_get(info, zone);
2006                         rcu_read_unlock();
2007
2008                         err = __ovs_ct_limit_get_zone_limit(
2009                                 net, info->data, zone, limit, reply);
2010                         if (err)
2011                                 return err;
2012                 }
2013                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2014                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2015                                 NLA_ALIGN(sizeof(*zone_limit)));
2016         }
2017
2018         if (rem)
2019                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2020
2021         return 0;
2022 }
2023
2024 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2025                                            struct ovs_ct_limit_info *info,
2026                                            struct sk_buff *reply)
2027 {
2028         struct ovs_ct_limit *ct_limit;
2029         struct hlist_head *head;
2030         int i, err = 0;
2031
2032         err = ovs_ct_limit_get_default_limit(info, reply);
2033         if (err)
2034                 return err;
2035
2036         rcu_read_lock();
2037         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2038                 head = &info->limits[i];
2039                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2040                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2041                                 ct_limit->zone, ct_limit->limit, reply);
2042                         if (err)
2043                                 goto exit_err;
2044                 }
2045         }
2046
2047 exit_err:
2048         rcu_read_unlock();
2049         return err;
2050 }
2051
2052 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2053 {
2054         struct nlattr **a = info->attrs;
2055         struct sk_buff *reply;
2056         struct ovs_header *ovs_reply_header;
2057         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2058         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2059         int err;
2060
2061         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2062                                              &ovs_reply_header);
2063         if (IS_ERR(reply))
2064                 return PTR_ERR(reply);
2065
2066         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2067                 err = -EINVAL;
2068                 goto exit_err;
2069         }
2070
2071         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2072                                           ct_limit_info);
2073         if (err)
2074                 goto exit_err;
2075
2076         static_branch_enable(&ovs_ct_limit_enabled);
2077
2078         genlmsg_end(reply, ovs_reply_header);
2079         return genlmsg_reply(reply, info);
2080
2081 exit_err:
2082         nlmsg_free(reply);
2083         return err;
2084 }
2085
2086 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2087 {
2088         struct nlattr **a = info->attrs;
2089         struct sk_buff *reply;
2090         struct ovs_header *ovs_reply_header;
2091         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2092         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2093         int err;
2094
2095         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2096                                              &ovs_reply_header);
2097         if (IS_ERR(reply))
2098                 return PTR_ERR(reply);
2099
2100         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2101                 err = -EINVAL;
2102                 goto exit_err;
2103         }
2104
2105         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2106                                           ct_limit_info);
2107         if (err)
2108                 goto exit_err;
2109
2110         genlmsg_end(reply, ovs_reply_header);
2111         return genlmsg_reply(reply, info);
2112
2113 exit_err:
2114         nlmsg_free(reply);
2115         return err;
2116 }
2117
2118 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2119 {
2120         struct nlattr **a = info->attrs;
2121         struct nlattr *nla_reply;
2122         struct sk_buff *reply;
2123         struct ovs_header *ovs_reply_header;
2124         struct net *net = sock_net(skb->sk);
2125         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2126         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2127         int err;
2128
2129         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2130                                              &ovs_reply_header);
2131         if (IS_ERR(reply))
2132                 return PTR_ERR(reply);
2133
2134         nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2135
2136         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2137                 err = ovs_ct_limit_get_zone_limit(
2138                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2139                         reply);
2140                 if (err)
2141                         goto exit_err;
2142         } else {
2143                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2144                                                       reply);
2145                 if (err)
2146                         goto exit_err;
2147         }
2148
2149         nla_nest_end(reply, nla_reply);
2150         genlmsg_end(reply, ovs_reply_header);
2151         return genlmsg_reply(reply, info);
2152
2153 exit_err:
2154         nlmsg_free(reply);
2155         return err;
2156 }
2157
2158 static struct genl_ops ct_limit_genl_ops[] = {
2159         { .cmd = OVS_CT_LIMIT_CMD_SET,
2160                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2161                                            * privilege. */
2162                 .policy = ct_limit_policy,
2163                 .doit = ovs_ct_limit_cmd_set,
2164         },
2165         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2166                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2167                                            * privilege. */
2168                 .policy = ct_limit_policy,
2169                 .doit = ovs_ct_limit_cmd_del,
2170         },
2171         { .cmd = OVS_CT_LIMIT_CMD_GET,
2172                 .flags = 0,               /* OK for unprivileged users. */
2173                 .policy = ct_limit_policy,
2174                 .doit = ovs_ct_limit_cmd_get,
2175         },
2176 };
2177
2178 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2179         .name = OVS_CT_LIMIT_MCGROUP,
2180 };
2181
2182 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2183         .hdrsize = sizeof(struct ovs_header),
2184         .name = OVS_CT_LIMIT_FAMILY,
2185         .version = OVS_CT_LIMIT_VERSION,
2186         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2187         .netnsok = true,
2188         .parallel_ops = true,
2189         .ops = ct_limit_genl_ops,
2190         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2191         .mcgrps = &ovs_ct_limit_multicast_group,
2192         .n_mcgrps = 1,
2193         .module = THIS_MODULE,
2194 };
2195 #endif
2196
2197 int ovs_ct_init(struct net *net)
2198 {
2199         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2200         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2201
2202         if (nf_connlabels_get(net, n_bits - 1)) {
2203                 ovs_net->xt_label = false;
2204                 OVS_NLERR(true, "Failed to set connlabel length");
2205         } else {
2206                 ovs_net->xt_label = true;
2207         }
2208
2209 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2210         return ovs_ct_limit_init(net, ovs_net);
2211 #else
2212         return 0;
2213 #endif
2214 }
2215
2216 void ovs_ct_exit(struct net *net)
2217 {
2218         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2219
2220 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2221         ovs_ct_limit_exit(net, ovs_net);
2222 #endif
2223
2224         if (ovs_net->xt_label)
2225                 nf_connlabels_put(net);
2226 }