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 #include <net/ipv6_frag.h>
  30
  31 #ifdef CONFIG_NF_NAT_NEEDED
  32 #include <linux/netfilter/nf_nat.h>
  33 #include <net/netfilter/nf_nat_core.h>
  34 #include <net/netfilter/nf_nat_l3proto.h>
  35 #endif
  36
  37 #include "datapath.h"
  38 #include "conntrack.h"
  39 #include "flow.h"
  40 #include "flow_netlink.h"
  41
  42 struct ovs_ct_len_tbl {
  43         int maxlen;
  44         int minlen;
  45 };
  46
  47 /* Metadata mark for masked write to conntrack mark */
  48 struct md_mark {
  49         u32 value;
  50         u32 mask;
  51 };
  52
  53 /* Metadata label for masked write to conntrack label. */
  54 struct md_labels {
  55         struct ovs_key_ct_labels value;
  56         struct ovs_key_ct_labels mask;
  57 };
  58
  59 enum ovs_ct_nat {
  60         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  61         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  62         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  63 };
  64
  65 /* Conntrack action context for execution. */
  66 struct ovs_conntrack_info {
  67         struct nf_conntrack_helper *helper;
  68         struct nf_conntrack_zone zone;
  69         struct nf_conn *ct;
  70         u8 commit : 1;
  71         u8 nat : 3;                 /* enum ovs_ct_nat */
  72         u8 force : 1;
  73         u8 have_eventmask : 1;
  74         u16 family;
  75         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  76         struct md_mark mark;
  77         struct md_labels labels;
  78 #ifdef CONFIG_NF_NAT_NEEDED
  79         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  80 #endif
  81 };
  82
  83 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  84 #define OVS_CT_LIMIT_UNLIMITED  0
  85 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  86 #define CT_LIMIT_HASH_BUCKETS 512
  87 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  88
  89 struct ovs_ct_limit {
  90         /* Elements in ovs_ct_limit_info->limits hash table */
  91         struct hlist_node hlist_node;
  92         struct rcu_head rcu;
  93         u16 zone;
  94         u32 limit;
  95 };
  96
  97 struct ovs_ct_limit_info {
  98         u32 default_limit;
  99         struct hlist_head *limits;
 100         struct nf_conncount_data *data;
 101 };
 102
 103 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
 104         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
 105 };
 106 #endif
 107
 108 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 109
 110 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 111
 112 static u16 key_to_nfproto(const struct sw_flow_key *key)
 113 {
 114         switch (ntohs(key->eth.type)) {
 115         case ETH_P_IP:
 116                 return NFPROTO_IPV4;
 117         case ETH_P_IPV6:
 118                 return NFPROTO_IPV6;
 119         default:
 120                 return NFPROTO_UNSPEC;
 121         }
 122 }
 123
 124 /* Map SKB connection state into the values used by flow definition. */
 125 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 126 {
 127         u8 ct_state = OVS_CS_F_TRACKED;
 128
 129         switch (ctinfo) {
 130         case IP_CT_ESTABLISHED_REPLY:
 131         case IP_CT_RELATED_REPLY:
 132                 ct_state |= OVS_CS_F_REPLY_DIR;
 133                 break;
 134         default:
 135                 break;
 136         }
 137
 138         switch (ctinfo) {
 139         case IP_CT_ESTABLISHED:
 140         case IP_CT_ESTABLISHED_REPLY:
 141                 ct_state |= OVS_CS_F_ESTABLISHED;
 142                 break;
 143         case IP_CT_RELATED:
 144         case IP_CT_RELATED_REPLY:
 145                 ct_state |= OVS_CS_F_RELATED;
 146                 break;
 147         case IP_CT_NEW:
 148                 ct_state |= OVS_CS_F_NEW;
 149                 break;
 150         default:
 151                 break;
 152         }
 153
 154         return ct_state;
 155 }
 156
 157 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 158 {
 159 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 160         return ct ? ct->mark : 0;
 161 #else
 162         return 0;
 163 #endif
 164 }
 165
 166 /* Guard against conntrack labels max size shrinking below 128 bits. */
 167 #if NF_CT_LABELS_MAX_SIZE < 16
 168 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 169 #endif
 170
 171 static void ovs_ct_get_labels(const struct nf_conn *ct,
 172                               struct ovs_key_ct_labels *labels)
 173 {
 174         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 175
 176         if (cl)
 177                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 178         else
 179                 memset(labels, 0, OVS_CT_LABELS_LEN);
 180 }
 181
 182 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 183                                         const struct nf_conntrack_tuple *orig,
 184                                         u8 icmp_proto)
 185 {
 186         key->ct_orig_proto = orig->dst.protonum;
 187         if (orig->dst.protonum == icmp_proto) {
 188                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 189                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 190         } else {
 191                 key->ct.orig_tp.src = orig->src.u.all;
 192                 key->ct.orig_tp.dst = orig->dst.u.all;
 193         }
 194 }
 195
 196 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 197                                 const struct nf_conntrack_zone *zone,
 198                                 const struct nf_conn *ct)
 199 {
 200         key->ct_state = state;
 201         key->ct_zone = zone->id;
 202         key->ct.mark = ovs_ct_get_mark(ct);
 203         ovs_ct_get_labels(ct, &key->ct.labels);
 204
 205         if (ct) {
 206                 const struct nf_conntrack_tuple *orig;
 207
 208                 /* Use the master if we have one. */
 209                 if (ct->master)
 210                         ct = ct->master;
 211                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 212
 213                 /* IP version must match with the master connection. */
 214                 if (key->eth.type == htons(ETH_P_IP) &&
 215                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 216                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 217                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 218                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 219                         return;
 220                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 221                            !sw_flow_key_is_nd(key) &&
 222                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 223                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 224                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 225                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 226                         return;
 227                 }
 228         }
 229         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 230          * original direction key fields.
 231          */
 232         key->ct_orig_proto = 0;
 233 }
 234
 235 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 236  * previously sent the packet to conntrack via the ct action.  If
 237  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 238  * initialized from the connection status.
 239  */
 240 static void ovs_ct_update_key(const struct sk_buff *skb,
 241                               const struct ovs_conntrack_info *info,
 242                               struct sw_flow_key *key, bool post_ct,
 243                               bool keep_nat_flags)
 244 {
 245         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 246         enum ip_conntrack_info ctinfo;
 247         struct nf_conn *ct;
 248         u8 state = 0;
 249
 250         ct = nf_ct_get(skb, &ctinfo);
 251         if (ct) {
 252                 state = ovs_ct_get_state(ctinfo);
 253                 /* All unconfirmed entries are NEW connections. */
 254                 if (!nf_ct_is_confirmed(ct))
 255                         state |= OVS_CS_F_NEW;
 256                 /* OVS persists the related flag for the duration of the
 257                  * connection.
 258                  */
 259                 if (ct->master)
 260                         state |= OVS_CS_F_RELATED;
 261                 if (keep_nat_flags) {
 262                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 263                 } else {
 264                         if (ct->status & IPS_SRC_NAT)
 265                                 state |= OVS_CS_F_SRC_NAT;
 266                         if (ct->status & IPS_DST_NAT)
 267                                 state |= OVS_CS_F_DST_NAT;
 268                 }
 269                 zone = nf_ct_zone(ct);
 270         } else if (post_ct) {
 271                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 272                 if (info)
 273                         zone = &info->zone;
 274         }
 275         __ovs_ct_update_key(key, state, zone, ct);
 276 }
 277
 278 /* This is called to initialize CT key fields possibly coming in from the local
 279  * stack.
 280  */
 281 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 282 {
 283         ovs_ct_update_key(skb, NULL, key, false, false);
 284 }
 285
 286 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 287                    const struct sw_flow_key *output, struct sk_buff *skb)
 288 {
 289         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 290                 return -EMSGSIZE;
 291
 292         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 293             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 294                 return -EMSGSIZE;
 295
 296         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 297             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 298                 return -EMSGSIZE;
 299
 300         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 301             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 302                     &output->ct.labels))
 303                 return -EMSGSIZE;
 304
 305         if (swkey->ct_orig_proto) {
 306                 if (swkey->eth.type == htons(ETH_P_IP)) {
 307                         struct ovs_key_ct_tuple_ipv4 orig;
 308
 309                         memset(&orig, 0, sizeof(orig));
 310                         orig.ipv4_src = output->ipv4.ct_orig.src;
 311                         orig.ipv4_dst = output->ipv4.ct_orig.dst;
 312                         orig.src_port = output->ct.orig_tp.src;
 313                         orig.dst_port = output->ct.orig_tp.dst;
 314                         orig.ipv4_proto = output->ct_orig_proto;
 315
 316                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 317                                     sizeof(orig), &orig))
 318                                 return -EMSGSIZE;
 319                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 320                         struct ovs_key_ct_tuple_ipv6 orig;
 321
 322                         memset(&orig, 0, sizeof(orig));
 323                         memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
 324                                sizeof(orig.ipv6_src));
 325                         memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
 326                                sizeof(orig.ipv6_dst));
 327                         orig.src_port = output->ct.orig_tp.src;
 328                         orig.dst_port = output->ct.orig_tp.dst;
 329                         orig.ipv6_proto = output->ct_orig_proto;
 330
 331                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 332                                     sizeof(orig), &orig))
 333                                 return -EMSGSIZE;
 334                 }
 335         }
 336
 337         return 0;
 338 }
 339
 340 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 341                            u32 ct_mark, u32 mask)
 342 {
 343 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 344         u32 new_mark;
 345
 346         new_mark = ct_mark | (ct->mark & ~(mask));
 347         if (ct->mark != new_mark) {
 348                 ct->mark = new_mark;
 349                 if (nf_ct_is_confirmed(ct))
 350                         nf_conntrack_event_cache(IPCT_MARK, ct);
 351                 key->ct.mark = new_mark;
 352         }
 353
 354         return 0;
 355 #else
 356         return -ENOTSUPP;
 357 #endif
 358 }
 359
 360 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 361 {
 362         struct nf_conn_labels *cl;
 363
 364         cl = nf_ct_labels_find(ct);
 365         if (!cl) {
 366                 nf_ct_labels_ext_add(ct);
 367                 cl = nf_ct_labels_find(ct);
 368         }
 369
 370         return cl;
 371 }
 372
 373 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 374  * since the new connection is not yet confirmed, and thus no-one else has
 375  * access to it's labels, we simply write them over.
 376  */
 377 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 378                               const struct ovs_key_ct_labels *labels,
 379                               const struct ovs_key_ct_labels *mask)
 380 {
 381         struct nf_conn_labels *cl, *master_cl;
 382         bool have_mask = labels_nonzero(mask);
 383
 384         /* Inherit master's labels to the related connection? */
 385         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 386
 387         if (!master_cl && !have_mask)
 388                 return 0;   /* Nothing to do. */
 389
 390         cl = ovs_ct_get_conn_labels(ct);
 391         if (!cl)
 392                 return -ENOSPC;
 393
 394         /* Inherit the master's labels, if any. */
 395         if (master_cl)
 396                 *cl = *master_cl;
 397
 398         if (have_mask) {
 399                 u32 *dst = (u32 *)cl->bits;
 400                 int i;
 401
 402                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 403                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 404                                 (labels->ct_labels_32[i]
 405                                  & mask->ct_labels_32[i]);
 406         }
 407
 408         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 409          * IPCT_LABEL bit is set in the event cache.
 410          */
 411         nf_conntrack_event_cache(IPCT_LABEL, ct);
 412
 413         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 414
 415         return 0;
 416 }
 417
 418 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 419                              const struct ovs_key_ct_labels *labels,
 420                              const struct ovs_key_ct_labels *mask)
 421 {
 422         struct nf_conn_labels *cl;
 423         int err;
 424
 425         cl = ovs_ct_get_conn_labels(ct);
 426         if (!cl)
 427                 return -ENOSPC;
 428
 429         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 430                                     mask->ct_labels_32,
 431                                     OVS_CT_LABELS_LEN_32);
 432         if (err)
 433                 return err;
 434
 435         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 436
 437         return 0;
 438 }
 439
 440 /* 'skb' should already be pulled to nh_ofs. */
 441 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 442 {
 443         const struct nf_conntrack_helper *helper;
 444         const struct nf_conn_help *help;
 445         enum ip_conntrack_info ctinfo;
 446         unsigned int protoff;
 447         struct nf_conn *ct;
 448         int err;
 449
 450         ct = nf_ct_get(skb, &ctinfo);
 451         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 452                 return NF_ACCEPT;
 453
 454         help = nfct_help(ct);
 455         if (!help)
 456                 return NF_ACCEPT;
 457
 458         helper = rcu_dereference(help->helper);
 459         if (!helper)
 460                 return NF_ACCEPT;
 461
 462         switch (proto) {
 463         case NFPROTO_IPV4:
 464                 protoff = ip_hdrlen(skb);
 465                 break;
 466         case NFPROTO_IPV6: {
 467                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 468                 __be16 frag_off;
 469                 int ofs;
 470
 471                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 472                                        &frag_off);
 473                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 474                         pr_debug("proto header not found\n");
 475                         return NF_ACCEPT;
 476                 }
 477                 protoff = ofs;
 478                 break;
 479         }
 480         default:
 481                 WARN_ONCE(1, "helper invoked on non-IP family!");
 482                 return NF_DROP;
 483         }
 484
 485         err = helper->help(skb, protoff, ct, ctinfo);
 486         if (err != NF_ACCEPT)
 487                 return err;
 488
 489         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 490          * FTP with NAT) adusting the TCP payload size when mangling IP
 491          * addresses and/or port numbers in the text-based control connection.
 492          */
 493         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 494             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 495                 return NF_DROP;
 496         return NF_ACCEPT;
 497 }
 498
 499 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 500  * value if 'skb' is freed.
 501  */
 502 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 503                             u16 zone, struct sk_buff *skb)
 504 {
 505         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 506         int err;
 507
 508         if (key->eth.type == htons(ETH_P_IP)) {
 509                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 510
 511                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 512                 err = ip_defrag(net, skb, user);
 513                 if (err)
 514                         return err;
 515
 516                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 517 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 518         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 519                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 520
 521                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 522                 err = nf_ct_frag6_gather(net, skb, user);
 523                 if (err) {
 524                         if (err != -EINPROGRESS)
 525                                 kfree_skb(skb);
 526                         return err;
 527                 }
 528
 529                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 530                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 531 #endif
 532         } else {
 533                 kfree_skb(skb);
 534                 return -EPFNOSUPPORT;
 535         }
 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_conntrack_put(&ct->ct_general);
 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_tuplepr(&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         /* Force conntrack entry direction to the current packet? */
 711         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 712                 /* Delete the conntrack entry if confirmed, else just release
 713                  * the reference.
 714                  */
 715                 if (nf_ct_is_confirmed(ct))
 716                         nf_ct_delete(ct, 0, 0);
 717
 718                 nf_conntrack_put(&ct->ct_general);
 719                 nf_ct_set(skb, NULL, 0);
 720                 return false;
 721         }
 722
 723         return ct_executed;
 724 }
 725
 726 #ifdef CONFIG_NF_NAT_NEEDED
 727 /* Modelled after nf_nat_ipv[46]_fn().
 728  * range is only used for new, uninitialized NAT state.
 729  * Returns either NF_ACCEPT or NF_DROP.
 730  */
 731 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 732                               enum ip_conntrack_info ctinfo,
 733                               const struct nf_nat_range2 *range,
 734                               enum nf_nat_manip_type maniptype)
 735 {
 736         int hooknum, nh_off, err = NF_ACCEPT;
 737
 738         nh_off = skb_network_offset(skb);
 739         skb_pull_rcsum(skb, nh_off);
 740
 741         /* See HOOK2MANIP(). */
 742         if (maniptype == NF_NAT_MANIP_SRC)
 743                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 744         else
 745                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 746
 747         switch (ctinfo) {
 748         case IP_CT_RELATED:
 749         case IP_CT_RELATED_REPLY:
 750                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 751                     skb->protocol == htons(ETH_P_IP) &&
 752                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 753                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 754                                                            hooknum))
 755                                 err = NF_DROP;
 756                         goto push;
 757                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 758                            skb->protocol == htons(ETH_P_IPV6)) {
 759                         __be16 frag_off;
 760                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 761                         int hdrlen = ipv6_skip_exthdr(skb,
 762                                                       sizeof(struct ipv6hdr),
 763                                                       &nexthdr, &frag_off);
 764
 765                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 766                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 767                                                                      ctinfo,
 768                                                                      hooknum,
 769                                                                      hdrlen))
 770                                         err = NF_DROP;
 771                                 goto push;
 772                         }
 773                 }
 774                 /* Non-ICMP, fall thru to initialize if needed. */
 775                 /* fall through */
 776         case IP_CT_NEW:
 777                 /* Seen it before?  This can happen for loopback, retrans,
 778                  * or local packets.
 779                  */
 780                 if (!nf_nat_initialized(ct, maniptype)) {
 781                         /* Initialize according to the NAT action. */
 782                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 783                                 /* Action is set up to establish a new
 784                                  * mapping.
 785                                  */
 786                                 ? nf_nat_setup_info(ct, range, maniptype)
 787                                 : nf_nat_alloc_null_binding(ct, hooknum);
 788                         if (err != NF_ACCEPT)
 789                                 goto push;
 790                 }
 791                 break;
 792
 793         case IP_CT_ESTABLISHED:
 794         case IP_CT_ESTABLISHED_REPLY:
 795                 break;
 796
 797         default:
 798                 err = NF_DROP;
 799                 goto push;
 800         }
 801
 802         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 803 push:
 804         skb_push(skb, nh_off);
 805         skb_postpush_rcsum(skb, skb->data, nh_off);
 806
 807         return err;
 808 }
 809
 810 static void ovs_nat_update_key(struct sw_flow_key *key,
 811                                const struct sk_buff *skb,
 812                                enum nf_nat_manip_type maniptype)
 813 {
 814         if (maniptype == NF_NAT_MANIP_SRC) {
 815                 __be16 src;
 816
 817                 key->ct_state |= OVS_CS_F_SRC_NAT;
 818                 if (key->eth.type == htons(ETH_P_IP))
 819                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 820                 else if (key->eth.type == htons(ETH_P_IPV6))
 821                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 822                                sizeof(key->ipv6.addr.src));
 823                 else
 824                         return;
 825
 826                 if (key->ip.proto == IPPROTO_UDP)
 827                         src = udp_hdr(skb)->source;
 828                 else if (key->ip.proto == IPPROTO_TCP)
 829                         src = tcp_hdr(skb)->source;
 830                 else if (key->ip.proto == IPPROTO_SCTP)
 831                         src = sctp_hdr(skb)->source;
 832                 else
 833                         return;
 834
 835                 key->tp.src = src;
 836         } else {
 837                 __be16 dst;
 838
 839                 key->ct_state |= OVS_CS_F_DST_NAT;
 840                 if (key->eth.type == htons(ETH_P_IP))
 841                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 842                 else if (key->eth.type == htons(ETH_P_IPV6))
 843                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 844                                sizeof(key->ipv6.addr.dst));
 845                 else
 846                         return;
 847
 848                 if (key->ip.proto == IPPROTO_UDP)
 849                         dst = udp_hdr(skb)->dest;
 850                 else if (key->ip.proto == IPPROTO_TCP)
 851                         dst = tcp_hdr(skb)->dest;
 852                 else if (key->ip.proto == IPPROTO_SCTP)
 853                         dst = sctp_hdr(skb)->dest;
 854                 else
 855                         return;
 856
 857                 key->tp.dst = dst;
 858         }
 859 }
 860
 861 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 862 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 863                       const struct ovs_conntrack_info *info,
 864                       struct sk_buff *skb, struct nf_conn *ct,
 865                       enum ip_conntrack_info ctinfo)
 866 {
 867         enum nf_nat_manip_type maniptype;
 868         int err;
 869
 870         /* Add NAT extension if not confirmed yet. */
 871         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 872                 return NF_ACCEPT;   /* Can't NAT. */
 873
 874         /* Determine NAT type.
 875          * Check if the NAT type can be deduced from the tracked connection.
 876          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 877          * when committing.
 878          */
 879         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 880             ct->status & IPS_NAT_MASK &&
 881             (ctinfo != IP_CT_RELATED || info->commit)) {
 882                 /* NAT an established or related connection like before. */
 883                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 884                         /* This is the REPLY direction for a connection
 885                          * for which NAT was applied in the forward
 886                          * direction.  Do the reverse NAT.
 887                          */
 888                         maniptype = ct->status & IPS_SRC_NAT
 889                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 890                 else
 891                         maniptype = ct->status & IPS_SRC_NAT
 892                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 893         } else if (info->nat & OVS_CT_SRC_NAT) {
 894                 maniptype = NF_NAT_MANIP_SRC;
 895         } else if (info->nat & OVS_CT_DST_NAT) {
 896                 maniptype = NF_NAT_MANIP_DST;
 897         } else {
 898                 return NF_ACCEPT; /* Connection is not NATed. */
 899         }
 900         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 901
 902         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
 903                 if (ct->status & IPS_SRC_NAT) {
 904                         if (maniptype == NF_NAT_MANIP_SRC)
 905                                 maniptype = NF_NAT_MANIP_DST;
 906                         else
 907                                 maniptype = NF_NAT_MANIP_SRC;
 908
 909                         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
 910                                                  maniptype);
 911                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 912                         err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
 913                                                  NF_NAT_MANIP_SRC);
 914                 }
 915         }
 916
 917         /* Mark NAT done if successful and update the flow key. */
 918         if (err == NF_ACCEPT)
 919                 ovs_nat_update_key(key, skb, maniptype);
 920
 921         return err;
 922 }
 923 #else /* !CONFIG_NF_NAT_NEEDED */
 924 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 925                       const struct ovs_conntrack_info *info,
 926                       struct sk_buff *skb, struct nf_conn *ct,
 927                       enum ip_conntrack_info ctinfo)
 928 {
 929         return NF_ACCEPT;
 930 }
 931 #endif
 932
 933 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 934  * not done already.  Update key with new CT state after passing the packet
 935  * through conntrack.
 936  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 937  * set to NULL and 0 will be returned.
 938  */
 939 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 940                            const struct ovs_conntrack_info *info,
 941                            struct sk_buff *skb)
 942 {
 943         /* If we are recirculating packets to match on conntrack fields and
 944          * committing with a separate conntrack action,  then we don't need to
 945          * actually run the packet through conntrack twice unless it's for a
 946          * different zone.
 947          */
 948         bool cached = skb_nfct_cached(net, key, info, skb);
 949         enum ip_conntrack_info ctinfo;
 950         struct nf_conn *ct;
 951
 952         if (!cached) {
 953                 struct nf_conn *tmpl = info->ct;
 954                 int err;
 955
 956                 /* Associate skb with specified zone. */
 957                 if (tmpl) {
 958                         if (skb_nfct(skb))
 959                                 nf_conntrack_put(skb_nfct(skb));
 960                         nf_conntrack_get(&tmpl->ct_general);
 961                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 962                 }
 963
 964                 err = nf_conntrack_in(net, info->family,
 965                                       NF_INET_PRE_ROUTING, skb);
 966                 if (err != NF_ACCEPT)
 967                         return -ENOENT;
 968
 969                 /* Clear CT state NAT flags to mark that we have not yet done
 970                  * NAT after the nf_conntrack_in() call.  We can actually clear
 971                  * the whole state, as it will be re-initialized below.
 972                  */
 973                 key->ct_state = 0;
 974
 975                 /* Update the key, but keep the NAT flags. */
 976                 ovs_ct_update_key(skb, info, key, true, true);
 977         }
 978
 979         ct = nf_ct_get(skb, &ctinfo);
 980         if (ct) {
 981                 /* Packets starting a new connection must be NATted before the
 982                  * helper, so that the helper knows about the NAT.  We enforce
 983                  * this by delaying both NAT and helper calls for unconfirmed
 984                  * connections until the committing CT action.  For later
 985                  * packets NAT and Helper may be called in either order.
 986                  *
 987                  * NAT will be done only if the CT action has NAT, and only
 988                  * once per packet (per zone), as guarded by the NAT bits in
 989                  * the key->ct_state.
 990                  */
 991                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 992                     (nf_ct_is_confirmed(ct) || info->commit) &&
 993                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 994                         return -EINVAL;
 995                 }
 996
 997                 /* Userspace may decide to perform a ct lookup without a helper
 998                  * specified followed by a (recirculate and) commit with one.
 999                  * Therefore, for unconfirmed connections which we will commit,
1000                  * we need to attach the helper here.
1001                  */
1002                 if (!nf_ct_is_confirmed(ct) && info->commit &&
1003                     info->helper && !nfct_help(ct)) {
1004                         int err = __nf_ct_try_assign_helper(ct, info->ct,
1005                                                             GFP_ATOMIC);
1006                         if (err)
1007                                 return err;
1008                 }
1009
1010                 /* Call the helper only if:
1011                  * - nf_conntrack_in() was executed above ("!cached") for a
1012                  *   confirmed connection, or
1013                  * - When committing an unconfirmed connection.
1014                  */
1015                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1016                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1017                         return -EINVAL;
1018                 }
1019         }
1020
1021         return 0;
1022 }
1023
1024 /* Lookup connection and read fields into key. */
1025 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1026                          const struct ovs_conntrack_info *info,
1027                          struct sk_buff *skb)
1028 {
1029         struct nf_conntrack_expect *exp;
1030
1031         /* If we pass an expected packet through nf_conntrack_in() the
1032          * expectation is typically removed, but the packet could still be
1033          * lost in upcall processing.  To prevent this from happening we
1034          * perform an explicit expectation lookup.  Expected connections are
1035          * always new, and will be passed through conntrack only when they are
1036          * committed, as it is OK to remove the expectation at that time.
1037          */
1038         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1039         if (exp) {
1040                 u8 state;
1041
1042                 /* NOTE: New connections are NATted and Helped only when
1043                  * committed, so we are not calling into NAT here.
1044                  */
1045                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1046                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1047         } else {
1048                 struct nf_conn *ct;
1049                 int err;
1050
1051                 err = __ovs_ct_lookup(net, key, info, skb);
1052                 if (err)
1053                         return err;
1054
1055                 ct = (struct nf_conn *)skb_nfct(skb);
1056                 if (ct)
1057                         nf_ct_deliver_cached_events(ct);
1058         }
1059
1060         return 0;
1061 }
1062
1063 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1064 {
1065         size_t i;
1066
1067         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1068                 if (labels->ct_labels_32[i])
1069                         return true;
1070
1071         return false;
1072 }
1073
1074 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1075 static struct hlist_head *ct_limit_hash_bucket(
1076         const struct ovs_ct_limit_info *info, u16 zone)
1077 {
1078         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1079 }
1080
1081 /* Call with ovs_mutex */
1082 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1083                          struct ovs_ct_limit *new_ct_limit)
1084 {
1085         struct ovs_ct_limit *ct_limit;
1086         struct hlist_head *head;
1087
1088         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1089         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1090                 if (ct_limit->zone == new_ct_limit->zone) {
1091                         hlist_replace_rcu(&ct_limit->hlist_node,
1092                                           &new_ct_limit->hlist_node);
1093                         kfree_rcu(ct_limit, rcu);
1094                         return;
1095                 }
1096         }
1097
1098         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1099 }
1100
1101 /* Call with ovs_mutex */
1102 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1103 {
1104         struct ovs_ct_limit *ct_limit;
1105         struct hlist_head *head;
1106         struct hlist_node *n;
1107
1108         head = ct_limit_hash_bucket(info, zone);
1109         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1110                 if (ct_limit->zone == zone) {
1111                         hlist_del_rcu(&ct_limit->hlist_node);
1112                         kfree_rcu(ct_limit, rcu);
1113                         return;
1114                 }
1115         }
1116 }
1117
1118 /* Call with RCU read lock */
1119 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1120 {
1121         struct ovs_ct_limit *ct_limit;
1122         struct hlist_head *head;
1123
1124         head = ct_limit_hash_bucket(info, zone);
1125         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1126                 if (ct_limit->zone == zone)
1127                         return ct_limit->limit;
1128         }
1129
1130         return info->default_limit;
1131 }
1132
1133 static int ovs_ct_check_limit(struct net *net,
1134                               const struct ovs_conntrack_info *info,
1135                               const struct nf_conntrack_tuple *tuple)
1136 {
1137         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1138         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1139         u32 per_zone_limit, connections;
1140         u32 conncount_key;
1141
1142         conncount_key = info->zone.id;
1143
1144         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1145         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1146                 return 0;
1147
1148         connections = nf_conncount_count(net, ct_limit_info->data,
1149                                          &conncount_key, tuple, &info->zone);
1150         if (connections > per_zone_limit)
1151                 return -ENOMEM;
1152
1153         return 0;
1154 }
1155 #endif
1156
1157 /* Lookup connection and confirm if unconfirmed. */
1158 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1159                          const struct ovs_conntrack_info *info,
1160                          struct sk_buff *skb)
1161 {
1162         enum ip_conntrack_info ctinfo;
1163         struct nf_conn *ct;
1164         int err;
1165
1166         err = __ovs_ct_lookup(net, key, info, skb);
1167         if (err)
1168                 return err;
1169
1170         /* The connection could be invalid, in which case this is a no-op.*/
1171         ct = nf_ct_get(skb, &ctinfo);
1172         if (!ct)
1173                 return 0;
1174
1175 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1176         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1177                 if (!nf_ct_is_confirmed(ct)) {
1178                         err = ovs_ct_check_limit(net, info,
1179                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1180                         if (err) {
1181                                 net_warn_ratelimited("openvswitch: zone: %u "
1182                                         "execeeds conntrack limit\n",
1183                                         info->zone.id);
1184                                 return err;
1185                         }
1186                 }
1187         }
1188 #endif
1189
1190         /* Set the conntrack event mask if given.  NEW and DELETE events have
1191          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1192          * typically would receive many kinds of updates.  Setting the event
1193          * mask allows those events to be filtered.  The set event mask will
1194          * remain in effect for the lifetime of the connection unless changed
1195          * by a further CT action with both the commit flag and the eventmask
1196          * option. */
1197         if (info->have_eventmask) {
1198                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1199
1200                 if (cache)
1201                         cache->ctmask = info->eventmask;
1202         }
1203
1204         /* Apply changes before confirming the connection so that the initial
1205          * conntrack NEW netlink event carries the values given in the CT
1206          * action.
1207          */
1208         if (info->mark.mask) {
1209                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1210                                       info->mark.mask);
1211                 if (err)
1212                         return err;
1213         }
1214         if (!nf_ct_is_confirmed(ct)) {
1215                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1216                                          &info->labels.mask);
1217                 if (err)
1218                         return err;
1219         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1220                    labels_nonzero(&info->labels.mask)) {
1221                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1222                                         &info->labels.mask);
1223                 if (err)
1224                         return err;
1225         }
1226         /* This will take care of sending queued events even if the connection
1227          * is already confirmed.
1228          */
1229         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1230                 return -EINVAL;
1231
1232         return 0;
1233 }
1234
1235 /* Trim the skb to the length specified by the IP/IPv6 header,
1236  * removing any trailing lower-layer padding. This prepares the skb
1237  * for higher-layer processing that assumes skb->len excludes padding
1238  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1239  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1240  */
1241 static int ovs_skb_network_trim(struct sk_buff *skb)
1242 {
1243         unsigned int len;
1244         int err;
1245
1246         switch (skb->protocol) {
1247         case htons(ETH_P_IP):
1248                 len = ntohs(ip_hdr(skb)->tot_len);
1249                 break;
1250         case htons(ETH_P_IPV6):
1251                 len = sizeof(struct ipv6hdr)
1252                         + ntohs(ipv6_hdr(skb)->payload_len);
1253                 break;
1254         default:
1255                 len = skb->len;
1256         }
1257
1258         err = pskb_trim_rcsum(skb, len);
1259         if (err)
1260                 kfree_skb(skb);
1261
1262         return err;
1263 }
1264
1265 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1266  * value if 'skb' is freed.
1267  */
1268 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1269                    struct sw_flow_key *key,
1270                    const struct ovs_conntrack_info *info)
1271 {
1272         int nh_ofs;
1273         int err;
1274
1275         /* The conntrack module expects to be working at L3. */
1276         nh_ofs = skb_network_offset(skb);
1277         skb_pull_rcsum(skb, nh_ofs);
1278
1279         err = ovs_skb_network_trim(skb);
1280         if (err)
1281                 return err;
1282
1283         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1284                 err = handle_fragments(net, key, info->zone.id, skb);
1285                 if (err)
1286                         return err;
1287         }
1288
1289         if (info->commit)
1290                 err = ovs_ct_commit(net, key, info, skb);
1291         else
1292                 err = ovs_ct_lookup(net, key, info, skb);
1293
1294         skb_push(skb, nh_ofs);
1295         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1296         if (err)
1297                 kfree_skb(skb);
1298         return err;
1299 }
1300
1301 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1302 {
1303         if (skb_nfct(skb)) {
1304                 nf_conntrack_put(skb_nfct(skb));
1305                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1306                 ovs_ct_fill_key(skb, key);
1307         }
1308
1309         return 0;
1310 }
1311
1312 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1313                              const struct sw_flow_key *key, bool log)
1314 {
1315         struct nf_conntrack_helper *helper;
1316         struct nf_conn_help *help;
1317
1318         helper = nf_conntrack_helper_try_module_get(name, info->family,
1319                                                     key->ip.proto);
1320         if (!helper) {
1321                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1322                 return -EINVAL;
1323         }
1324
1325         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1326         if (!help) {
1327                 nf_conntrack_helper_put(helper);
1328                 return -ENOMEM;
1329         }
1330
1331         rcu_assign_pointer(help->helper, helper);
1332         info->helper = helper;
1333
1334         if (info->nat)
1335                 request_module("ip_nat_%s", name);
1336
1337         return 0;
1338 }
1339
1340 #ifdef CONFIG_NF_NAT_NEEDED
1341 static int parse_nat(const struct nlattr *attr,
1342                      struct ovs_conntrack_info *info, bool log)
1343 {
1344         struct nlattr *a;
1345         int rem;
1346         bool have_ip_max = false;
1347         bool have_proto_max = false;
1348         bool ip_vers = (info->family == NFPROTO_IPV6);
1349
1350         nla_for_each_nested(a, attr, rem) {
1351                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1352                         [OVS_NAT_ATTR_SRC] = {0, 0},
1353                         [OVS_NAT_ATTR_DST] = {0, 0},
1354                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1355                                                  sizeof(struct in6_addr)},
1356                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1357                                                  sizeof(struct in6_addr)},
1358                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1359                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1360                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1361                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1362                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1363                 };
1364                 int type = nla_type(a);
1365
1366                 if (type > OVS_NAT_ATTR_MAX) {
1367                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1368                                   type, OVS_NAT_ATTR_MAX);
1369                         return -EINVAL;
1370                 }
1371
1372                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1373                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1374                                   type, nla_len(a),
1375                                   ovs_nat_attr_lens[type][ip_vers]);
1376                         return -EINVAL;
1377                 }
1378
1379                 switch (type) {
1380                 case OVS_NAT_ATTR_SRC:
1381                 case OVS_NAT_ATTR_DST:
1382                         if (info->nat) {
1383                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1384                                 return -ERANGE;
1385                         }
1386                         info->nat |= OVS_CT_NAT;
1387                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1388                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1389                         break;
1390
1391                 case OVS_NAT_ATTR_IP_MIN:
1392                         nla_memcpy(&info->range.min_addr, a,
1393                                    sizeof(info->range.min_addr));
1394                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1395                         break;
1396
1397                 case OVS_NAT_ATTR_IP_MAX:
1398                         have_ip_max = true;
1399                         nla_memcpy(&info->range.max_addr, a,
1400                                    sizeof(info->range.max_addr));
1401                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1402                         break;
1403
1404                 case OVS_NAT_ATTR_PROTO_MIN:
1405                         info->range.min_proto.all = htons(nla_get_u16(a));
1406                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1407                         break;
1408
1409                 case OVS_NAT_ATTR_PROTO_MAX:
1410                         have_proto_max = true;
1411                         info->range.max_proto.all = htons(nla_get_u16(a));
1412                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1413                         break;
1414
1415                 case OVS_NAT_ATTR_PERSISTENT:
1416                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1417                         break;
1418
1419                 case OVS_NAT_ATTR_PROTO_HASH:
1420                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1421                         break;
1422
1423                 case OVS_NAT_ATTR_PROTO_RANDOM:
1424                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1425                         break;
1426
1427                 default:
1428                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1429                         return -EINVAL;
1430                 }
1431         }
1432
1433         if (rem > 0) {
1434                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1435                 return -EINVAL;
1436         }
1437         if (!info->nat) {
1438                 /* Do not allow flags if no type is given. */
1439                 if (info->range.flags) {
1440                         OVS_NLERR(log,
1441                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1442                                   );
1443                         return -EINVAL;
1444                 }
1445                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1446         } else if (!info->commit) {
1447                 OVS_NLERR(log,
1448                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1449                           );
1450                 return -EINVAL;
1451         }
1452         /* Allow missing IP_MAX. */
1453         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1454                 memcpy(&info->range.max_addr, &info->range.min_addr,
1455                        sizeof(info->range.max_addr));
1456         }
1457         /* Allow missing PROTO_MAX. */
1458         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1459             !have_proto_max) {
1460                 info->range.max_proto.all = info->range.min_proto.all;
1461         }
1462         return 0;
1463 }
1464 #endif
1465
1466 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1467         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1468         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1469         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1470                                     .maxlen = sizeof(u16) },
1471         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1472                                     .maxlen = sizeof(struct md_mark) },
1473         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1474                                     .maxlen = sizeof(struct md_labels) },
1475         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1476                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1477 #ifdef CONFIG_NF_NAT_NEEDED
1478         /* NAT length is checked when parsing the nested attributes. */
1479         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1480 #endif
1481         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1482                                     .maxlen = sizeof(u32) },
1483 };
1484
1485 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1486                     const char **helper, bool log)
1487 {
1488         struct nlattr *a;
1489         int rem;
1490
1491         nla_for_each_nested(a, attr, rem) {
1492                 int type = nla_type(a);
1493                 int maxlen;
1494                 int minlen;
1495
1496                 if (type > OVS_CT_ATTR_MAX) {
1497                         OVS_NLERR(log,
1498                                   "Unknown conntrack attr (type=%d, max=%d)",
1499                                   type, OVS_CT_ATTR_MAX);
1500                         return -EINVAL;
1501                 }
1502
1503                 maxlen = ovs_ct_attr_lens[type].maxlen;
1504                 minlen = ovs_ct_attr_lens[type].minlen;
1505                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1506                         OVS_NLERR(log,
1507                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1508                                   type, nla_len(a), maxlen);
1509                         return -EINVAL;
1510                 }
1511
1512                 switch (type) {
1513                 case OVS_CT_ATTR_FORCE_COMMIT:
1514                         info->force = true;
1515                         /* fall through. */
1516                 case OVS_CT_ATTR_COMMIT:
1517                         info->commit = true;
1518                         break;
1519 #ifdef CONFIG_NF_CONNTRACK_ZONES
1520                 case OVS_CT_ATTR_ZONE:
1521                         info->zone.id = nla_get_u16(a);
1522                         break;
1523 #endif
1524 #ifdef CONFIG_NF_CONNTRACK_MARK
1525                 case OVS_CT_ATTR_MARK: {
1526                         struct md_mark *mark = nla_data(a);
1527
1528                         if (!mark->mask) {
1529                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1530                                 return -EINVAL;
1531                         }
1532                         info->mark = *mark;
1533                         break;
1534                 }
1535 #endif
1536 #ifdef CONFIG_NF_CONNTRACK_LABELS
1537                 case OVS_CT_ATTR_LABELS: {
1538                         struct md_labels *labels = nla_data(a);
1539
1540                         if (!labels_nonzero(&labels->mask)) {
1541                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1542                                 return -EINVAL;
1543                         }
1544                         info->labels = *labels;
1545                         break;
1546                 }
1547 #endif
1548                 case OVS_CT_ATTR_HELPER:
1549                         *helper = nla_data(a);
1550                         if (!memchr(*helper, '\0', nla_len(a))) {
1551                                 OVS_NLERR(log, "Invalid conntrack helper");
1552                                 return -EINVAL;
1553                         }
1554                         break;
1555 #ifdef CONFIG_NF_NAT_NEEDED
1556                 case OVS_CT_ATTR_NAT: {
1557                         int err = parse_nat(a, info, log);
1558
1559                         if (err)
1560                                 return err;
1561                         break;
1562                 }
1563 #endif
1564                 case OVS_CT_ATTR_EVENTMASK:
1565                         info->have_eventmask = true;
1566                         info->eventmask = nla_get_u32(a);
1567                         break;
1568
1569                 default:
1570                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1571                                   type);
1572                         return -EINVAL;
1573                 }
1574         }
1575
1576 #ifdef CONFIG_NF_CONNTRACK_MARK
1577         if (!info->commit && info->mark.mask) {
1578                 OVS_NLERR(log,
1579                           "Setting conntrack mark requires 'commit' flag.");
1580                 return -EINVAL;
1581         }
1582 #endif
1583 #ifdef CONFIG_NF_CONNTRACK_LABELS
1584         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1585                 OVS_NLERR(log,
1586                           "Setting conntrack labels requires 'commit' flag.");
1587                 return -EINVAL;
1588         }
1589 #endif
1590         if (rem > 0) {
1591                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1592                 return -EINVAL;
1593         }
1594
1595         return 0;
1596 }
1597
1598 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1599 {
1600         if (attr == OVS_KEY_ATTR_CT_STATE)
1601                 return true;
1602         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1603             attr == OVS_KEY_ATTR_CT_ZONE)
1604                 return true;
1605         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1606             attr == OVS_KEY_ATTR_CT_MARK)
1607                 return true;
1608         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1609             attr == OVS_KEY_ATTR_CT_LABELS) {
1610                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1611
1612                 return ovs_net->xt_label;
1613         }
1614
1615         return false;
1616 }
1617
1618 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1619                        const struct sw_flow_key *key,
1620                        struct sw_flow_actions **sfa,  bool log)
1621 {
1622         struct ovs_conntrack_info ct_info;
1623         const char *helper = NULL;
1624         u16 family;
1625         int err;
1626
1627         family = key_to_nfproto(key);
1628         if (family == NFPROTO_UNSPEC) {
1629                 OVS_NLERR(log, "ct family unspecified");
1630                 return -EINVAL;
1631         }
1632
1633         memset(&ct_info, 0, sizeof(ct_info));
1634         ct_info.family = family;
1635
1636         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1637                         NF_CT_DEFAULT_ZONE_DIR, 0);
1638
1639         err = parse_ct(attr, &ct_info, &helper, log);
1640         if (err)
1641                 return err;
1642
1643         /* Set up template for tracking connections in specific zones. */
1644         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1645         if (!ct_info.ct) {
1646                 OVS_NLERR(log, "Failed to allocate conntrack template");
1647                 return -ENOMEM;
1648         }
1649         if (helper) {
1650                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1651                 if (err)
1652                         goto err_free_ct;
1653         }
1654
1655         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1656                                  sizeof(ct_info), log);
1657         if (err)
1658                 goto err_free_ct;
1659
1660         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1661         nf_conntrack_get(&ct_info.ct->ct_general);
1662         return 0;
1663 err_free_ct:
1664         __ovs_ct_free_action(&ct_info);
1665         return err;
1666 }
1667
1668 #ifdef CONFIG_NF_NAT_NEEDED
1669 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1670                                struct sk_buff *skb)
1671 {
1672         struct nlattr *start;
1673
1674         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1675         if (!start)
1676                 return false;
1677
1678         if (info->nat & OVS_CT_SRC_NAT) {
1679                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1680                         return false;
1681         } else if (info->nat & OVS_CT_DST_NAT) {
1682                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1683                         return false;
1684         } else {
1685                 goto out;
1686         }
1687
1688         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1689                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1690                     info->family == NFPROTO_IPV4) {
1691                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1692                                             info->range.min_addr.ip) ||
1693                             (info->range.max_addr.ip
1694                              != info->range.min_addr.ip &&
1695                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1696                                               info->range.max_addr.ip))))
1697                                 return false;
1698                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1699                            info->family == NFPROTO_IPV6) {
1700                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1701                                              &info->range.min_addr.in6) ||
1702                             (memcmp(&info->range.max_addr.in6,
1703                                     &info->range.min_addr.in6,
1704                                     sizeof(info->range.max_addr.in6)) &&
1705                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1706                                                &info->range.max_addr.in6))))
1707                                 return false;
1708                 } else {
1709                         return false;
1710                 }
1711         }
1712         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1713             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1714                          ntohs(info->range.min_proto.all)) ||
1715              (info->range.max_proto.all != info->range.min_proto.all &&
1716               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1717                           ntohs(info->range.max_proto.all)))))
1718                 return false;
1719
1720         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1721             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1722                 return false;
1723         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1724             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1725                 return false;
1726         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1727             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1728                 return false;
1729 out:
1730         nla_nest_end(skb, start);
1731
1732         return true;
1733 }
1734 #endif
1735
1736 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1737                           struct sk_buff *skb)
1738 {
1739         struct nlattr *start;
1740
1741         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1742         if (!start)
1743                 return -EMSGSIZE;
1744
1745         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1746                                             ? OVS_CT_ATTR_FORCE_COMMIT
1747                                             : OVS_CT_ATTR_COMMIT))
1748                 return -EMSGSIZE;
1749         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1750             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1751                 return -EMSGSIZE;
1752         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1753             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1754                     &ct_info->mark))
1755                 return -EMSGSIZE;
1756         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1757             labels_nonzero(&ct_info->labels.mask) &&
1758             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1759                     &ct_info->labels))
1760                 return -EMSGSIZE;
1761         if (ct_info->helper) {
1762                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1763                                    ct_info->helper->name))
1764                         return -EMSGSIZE;
1765         }
1766         if (ct_info->have_eventmask &&
1767             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1768                 return -EMSGSIZE;
1769
1770 #ifdef CONFIG_NF_NAT_NEEDED
1771         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1772                 return -EMSGSIZE;
1773 #endif
1774         nla_nest_end(skb, start);
1775
1776         return 0;
1777 }
1778
1779 void ovs_ct_free_action(const struct nlattr *a)
1780 {
1781         struct ovs_conntrack_info *ct_info = nla_data(a);
1782
1783         __ovs_ct_free_action(ct_info);
1784 }
1785
1786 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1787 {
1788         if (ct_info->helper)
1789                 nf_conntrack_helper_put(ct_info->helper);
1790         if (ct_info->ct)
1791                 nf_ct_tmpl_free(ct_info->ct);
1792 }
1793
1794 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1795 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1796 {
1797         int i, err;
1798
1799         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1800                                          GFP_KERNEL);
1801         if (!ovs_net->ct_limit_info)
1802                 return -ENOMEM;
1803
1804         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1805         ovs_net->ct_limit_info->limits =
1806                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1807                               GFP_KERNEL);
1808         if (!ovs_net->ct_limit_info->limits) {
1809                 kfree(ovs_net->ct_limit_info);
1810                 return -ENOMEM;
1811         }
1812
1813         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1814                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1815
1816         ovs_net->ct_limit_info->data =
1817                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1818
1819         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1820                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1821                 kfree(ovs_net->ct_limit_info->limits);
1822                 kfree(ovs_net->ct_limit_info);
1823                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1824                 return err;
1825         }
1826         return 0;
1827 }
1828
1829 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1830 {
1831         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1832         int i;
1833
1834         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1835         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1836                 struct hlist_head *head = &info->limits[i];
1837                 struct ovs_ct_limit *ct_limit;
1838
1839                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1840                         kfree_rcu(ct_limit, rcu);
1841         }
1842         kfree(ovs_net->ct_limit_info->limits);
1843         kfree(ovs_net->ct_limit_info);
1844 }
1845
1846 static struct sk_buff *
1847 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1848                              struct ovs_header **ovs_reply_header)
1849 {
1850         struct ovs_header *ovs_header = info->userhdr;
1851         struct sk_buff *skb;
1852
1853         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1854         if (!skb)
1855                 return ERR_PTR(-ENOMEM);
1856
1857         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1858                                         info->snd_seq,
1859                                         &dp_ct_limit_genl_family, 0, cmd);
1860
1861         if (!*ovs_reply_header) {
1862                 nlmsg_free(skb);
1863                 return ERR_PTR(-EMSGSIZE);
1864         }
1865         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1866
1867         return skb;
1868 }
1869
1870 static bool check_zone_id(int zone_id, u16 *pzone)
1871 {
1872         if (zone_id >= 0 && zone_id <= 65535) {
1873                 *pzone = (u16)zone_id;
1874                 return true;
1875         }
1876         return false;
1877 }
1878
1879 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1880                                        struct ovs_ct_limit_info *info)
1881 {
1882         struct ovs_zone_limit *zone_limit;
1883         int rem;
1884         u16 zone;
1885
1886         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1887         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1888
1889         while (rem >= sizeof(*zone_limit)) {
1890                 if (unlikely(zone_limit->zone_id ==
1891                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1892                         ovs_lock();
1893                         info->default_limit = zone_limit->limit;
1894                         ovs_unlock();
1895                 } else if (unlikely(!check_zone_id(
1896                                 zone_limit->zone_id, &zone))) {
1897                         OVS_NLERR(true, "zone id is out of range");
1898                 } else {
1899                         struct ovs_ct_limit *ct_limit;
1900
1901                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1902                         if (!ct_limit)
1903                                 return -ENOMEM;
1904
1905                         ct_limit->zone = zone;
1906                         ct_limit->limit = zone_limit->limit;
1907
1908                         ovs_lock();
1909                         ct_limit_set(info, ct_limit);
1910                         ovs_unlock();
1911                 }
1912                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1913                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1914                                 NLA_ALIGN(sizeof(*zone_limit)));
1915         }
1916
1917         if (rem)
1918                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1919
1920         return 0;
1921 }
1922
1923 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1924                                        struct ovs_ct_limit_info *info)
1925 {
1926         struct ovs_zone_limit *zone_limit;
1927         int rem;
1928         u16 zone;
1929
1930         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1931         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1932
1933         while (rem >= sizeof(*zone_limit)) {
1934                 if (unlikely(zone_limit->zone_id ==
1935                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1936                         ovs_lock();
1937                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1938                         ovs_unlock();
1939                 } else if (unlikely(!check_zone_id(
1940                                 zone_limit->zone_id, &zone))) {
1941                         OVS_NLERR(true, "zone id is out of range");
1942                 } else {
1943                         ovs_lock();
1944                         ct_limit_del(info, zone);
1945                         ovs_unlock();
1946                 }
1947                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1948                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1949                                 NLA_ALIGN(sizeof(*zone_limit)));
1950         }
1951
1952         if (rem)
1953                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1954
1955         return 0;
1956 }
1957
1958 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1959                                           struct sk_buff *reply)
1960 {
1961         struct ovs_zone_limit zone_limit;
1962         int err;
1963
1964         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1965         zone_limit.limit = info->default_limit;
1966         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1967         if (err)
1968                 return err;
1969
1970         return 0;
1971 }
1972
1973 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1974                                          struct nf_conncount_data *data,
1975                                          u16 zone_id, u32 limit,
1976                                          struct sk_buff *reply)
1977 {
1978         struct nf_conntrack_zone ct_zone;
1979         struct ovs_zone_limit zone_limit;
1980         u32 conncount_key = zone_id;
1981
1982         zone_limit.zone_id = zone_id;
1983         zone_limit.limit = limit;
1984         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1985
1986         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1987                                               &ct_zone);
1988         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1989 }
1990
1991 static int ovs_ct_limit_get_zone_limit(struct net *net,
1992                                        struct nlattr *nla_zone_limit,
1993                                        struct ovs_ct_limit_info *info,
1994                                        struct sk_buff *reply)
1995 {
1996         struct ovs_zone_limit *zone_limit;
1997         int rem, err;
1998         u32 limit;
1999         u16 zone;
2000
2001         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2002         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2003
2004         while (rem >= sizeof(*zone_limit)) {
2005                 if (unlikely(zone_limit->zone_id ==
2006                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2007                         err = ovs_ct_limit_get_default_limit(info, reply);
2008                         if (err)
2009                                 return err;
2010                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2011                                                         &zone))) {
2012                         OVS_NLERR(true, "zone id is out of range");
2013                 } else {
2014                         rcu_read_lock();
2015                         limit = ct_limit_get(info, zone);
2016                         rcu_read_unlock();
2017
2018                         err = __ovs_ct_limit_get_zone_limit(
2019                                 net, info->data, zone, limit, reply);
2020                         if (err)
2021                                 return err;
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, "get zone limit has %d unknown bytes", rem);
2030
2031         return 0;
2032 }
2033
2034 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2035                                            struct ovs_ct_limit_info *info,
2036                                            struct sk_buff *reply)
2037 {
2038         struct ovs_ct_limit *ct_limit;
2039         struct hlist_head *head;
2040         int i, err = 0;
2041
2042         err = ovs_ct_limit_get_default_limit(info, reply);
2043         if (err)
2044                 return err;
2045
2046         rcu_read_lock();
2047         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2048                 head = &info->limits[i];
2049                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2050                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2051                                 ct_limit->zone, ct_limit->limit, reply);
2052                         if (err)
2053                                 goto exit_err;
2054                 }
2055         }
2056
2057 exit_err:
2058         rcu_read_unlock();
2059         return err;
2060 }
2061
2062 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2063 {
2064         struct nlattr **a = info->attrs;
2065         struct sk_buff *reply;
2066         struct ovs_header *ovs_reply_header;
2067         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2068         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2069         int err;
2070
2071         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2072                                              &ovs_reply_header);
2073         if (IS_ERR(reply))
2074                 return PTR_ERR(reply);
2075
2076         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2077                 err = -EINVAL;
2078                 goto exit_err;
2079         }
2080
2081         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2082                                           ct_limit_info);
2083         if (err)
2084                 goto exit_err;
2085
2086         static_branch_enable(&ovs_ct_limit_enabled);
2087
2088         genlmsg_end(reply, ovs_reply_header);
2089         return genlmsg_reply(reply, info);
2090
2091 exit_err:
2092         nlmsg_free(reply);
2093         return err;
2094 }
2095
2096 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2097 {
2098         struct nlattr **a = info->attrs;
2099         struct sk_buff *reply;
2100         struct ovs_header *ovs_reply_header;
2101         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2102         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2103         int err;
2104
2105         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2106                                              &ovs_reply_header);
2107         if (IS_ERR(reply))
2108                 return PTR_ERR(reply);
2109
2110         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2111                 err = -EINVAL;
2112                 goto exit_err;
2113         }
2114
2115         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2116                                           ct_limit_info);
2117         if (err)
2118                 goto exit_err;
2119
2120         genlmsg_end(reply, ovs_reply_header);
2121         return genlmsg_reply(reply, info);
2122
2123 exit_err:
2124         nlmsg_free(reply);
2125         return err;
2126 }
2127
2128 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2129 {
2130         struct nlattr **a = info->attrs;
2131         struct nlattr *nla_reply;
2132         struct sk_buff *reply;
2133         struct ovs_header *ovs_reply_header;
2134         struct net *net = sock_net(skb->sk);
2135         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2136         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2137         int err;
2138
2139         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2140                                              &ovs_reply_header);
2141         if (IS_ERR(reply))
2142                 return PTR_ERR(reply);
2143
2144         nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2145
2146         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2147                 err = ovs_ct_limit_get_zone_limit(
2148                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2149                         reply);
2150                 if (err)
2151                         goto exit_err;
2152         } else {
2153                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2154                                                       reply);
2155                 if (err)
2156                         goto exit_err;
2157         }
2158
2159         nla_nest_end(reply, nla_reply);
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 struct genl_ops ct_limit_genl_ops[] = {
2169         { .cmd = OVS_CT_LIMIT_CMD_SET,
2170                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2171                                            * privilege. */
2172                 .policy = ct_limit_policy,
2173                 .doit = ovs_ct_limit_cmd_set,
2174         },
2175         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2176                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2177                                            * privilege. */
2178                 .policy = ct_limit_policy,
2179                 .doit = ovs_ct_limit_cmd_del,
2180         },
2181         { .cmd = OVS_CT_LIMIT_CMD_GET,
2182                 .flags = 0,               /* OK for unprivileged users. */
2183                 .policy = ct_limit_policy,
2184                 .doit = ovs_ct_limit_cmd_get,
2185         },
2186 };
2187
2188 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2189         .name = OVS_CT_LIMIT_MCGROUP,
2190 };
2191
2192 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2193         .hdrsize = sizeof(struct ovs_header),
2194         .name = OVS_CT_LIMIT_FAMILY,
2195         .version = OVS_CT_LIMIT_VERSION,
2196         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2197         .netnsok = true,
2198         .parallel_ops = true,
2199         .ops = ct_limit_genl_ops,
2200         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2201         .mcgrps = &ovs_ct_limit_multicast_group,
2202         .n_mcgrps = 1,
2203         .module = THIS_MODULE,
2204 };
2205 #endif
2206
2207 int ovs_ct_init(struct net *net)
2208 {
2209         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2210         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2211
2212         if (nf_connlabels_get(net, n_bits - 1)) {
2213                 ovs_net->xt_label = false;
2214                 OVS_NLERR(true, "Failed to set connlabel length");
2215         } else {
2216                 ovs_net->xt_label = true;
2217         }
2218
2219 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2220         return ovs_ct_limit_init(net, ovs_net);
2221 #else
2222         return 0;
2223 #endif
2224 }
2225
2226 void ovs_ct_exit(struct net *net)
2227 {
2228         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2229
2230 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2231         ovs_ct_limit_exit(net, ovs_net);
2232 #endif
2233
2234         if (ovs_net->xt_label)
2235                 nf_connlabels_put(net);
2236 }