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 #define IN6_ADDR_INITIALIZER(ADDR) \
 287         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 288           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 289
 290 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 291                    const struct sw_flow_key *output, struct sk_buff *skb)
 292 {
 293         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 294                 return -EMSGSIZE;
 295
 296         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 297             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 298                 return -EMSGSIZE;
 299
 300         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 301             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 302                 return -EMSGSIZE;
 303
 304         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 305             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 306                     &output->ct.labels))
 307                 return -EMSGSIZE;
 308
 309         if (swkey->ct_orig_proto) {
 310                 if (swkey->eth.type == htons(ETH_P_IP)) {
 311                         struct ovs_key_ct_tuple_ipv4 orig = {
 312                                 output->ipv4.ct_orig.src,
 313                                 output->ipv4.ct_orig.dst,
 314                                 output->ct.orig_tp.src,
 315                                 output->ct.orig_tp.dst,
 316                                 output->ct_orig_proto,
 317                         };
 318                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 319                                     sizeof(orig), &orig))
 320                                 return -EMSGSIZE;
 321                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 322                         struct ovs_key_ct_tuple_ipv6 orig = {
 323                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 324                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 325                                 output->ct.orig_tp.src,
 326                                 output->ct.orig_tp.dst,
 327                                 output->ct_orig_proto,
 328                         };
 329                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 330                                     sizeof(orig), &orig))
 331                                 return -EMSGSIZE;
 332                 }
 333         }
 334
 335         return 0;
 336 }
 337
 338 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 339                            u32 ct_mark, u32 mask)
 340 {
 341 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 342         u32 new_mark;
 343
 344         new_mark = ct_mark | (ct->mark & ~(mask));
 345         if (ct->mark != new_mark) {
 346                 ct->mark = new_mark;
 347                 if (nf_ct_is_confirmed(ct))
 348                         nf_conntrack_event_cache(IPCT_MARK, ct);
 349                 key->ct.mark = new_mark;
 350         }
 351
 352         return 0;
 353 #else
 354         return -ENOTSUPP;
 355 #endif
 356 }
 357
 358 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 359 {
 360         struct nf_conn_labels *cl;
 361
 362         cl = nf_ct_labels_find(ct);
 363         if (!cl) {
 364                 nf_ct_labels_ext_add(ct);
 365                 cl = nf_ct_labels_find(ct);
 366         }
 367
 368         return cl;
 369 }
 370
 371 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 372  * since the new connection is not yet confirmed, and thus no-one else has
 373  * access to it's labels, we simply write them over.
 374  */
 375 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 376                               const struct ovs_key_ct_labels *labels,
 377                               const struct ovs_key_ct_labels *mask)
 378 {
 379         struct nf_conn_labels *cl, *master_cl;
 380         bool have_mask = labels_nonzero(mask);
 381
 382         /* Inherit master's labels to the related connection? */
 383         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 384
 385         if (!master_cl && !have_mask)
 386                 return 0;   /* Nothing to do. */
 387
 388         cl = ovs_ct_get_conn_labels(ct);
 389         if (!cl)
 390                 return -ENOSPC;
 391
 392         /* Inherit the master's labels, if any. */
 393         if (master_cl)
 394                 *cl = *master_cl;
 395
 396         if (have_mask) {
 397                 u32 *dst = (u32 *)cl->bits;
 398                 int i;
 399
 400                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 401                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 402                                 (labels->ct_labels_32[i]
 403                                  & mask->ct_labels_32[i]);
 404         }
 405
 406         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 407          * IPCT_LABEL bit is set in the event cache.
 408          */
 409         nf_conntrack_event_cache(IPCT_LABEL, ct);
 410
 411         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 412
 413         return 0;
 414 }
 415
 416 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 417                              const struct ovs_key_ct_labels *labels,
 418                              const struct ovs_key_ct_labels *mask)
 419 {
 420         struct nf_conn_labels *cl;
 421         int err;
 422
 423         cl = ovs_ct_get_conn_labels(ct);
 424         if (!cl)
 425                 return -ENOSPC;
 426
 427         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 428                                     mask->ct_labels_32,
 429                                     OVS_CT_LABELS_LEN_32);
 430         if (err)
 431                 return err;
 432
 433         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 434
 435         return 0;
 436 }
 437
 438 /* 'skb' should already be pulled to nh_ofs. */
 439 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 440 {
 441         const struct nf_conntrack_helper *helper;
 442         const struct nf_conn_help *help;
 443         enum ip_conntrack_info ctinfo;
 444         unsigned int protoff;
 445         struct nf_conn *ct;
 446         int err;
 447
 448         ct = nf_ct_get(skb, &ctinfo);
 449         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 450                 return NF_ACCEPT;
 451
 452         help = nfct_help(ct);
 453         if (!help)
 454                 return NF_ACCEPT;
 455
 456         helper = rcu_dereference(help->helper);
 457         if (!helper)
 458                 return NF_ACCEPT;
 459
 460         switch (proto) {
 461         case NFPROTO_IPV4:
 462                 protoff = ip_hdrlen(skb);
 463                 break;
 464         case NFPROTO_IPV6: {
 465                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 466                 __be16 frag_off;
 467                 int ofs;
 468
 469                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 470                                        &frag_off);
 471                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 472                         pr_debug("proto header not found\n");
 473                         return NF_ACCEPT;
 474                 }
 475                 protoff = ofs;
 476                 break;
 477         }
 478         default:
 479                 WARN_ONCE(1, "helper invoked on non-IP family!");
 480                 return NF_DROP;
 481         }
 482
 483         err = helper->help(skb, protoff, ct, ctinfo);
 484         if (err != NF_ACCEPT)
 485                 return err;
 486
 487         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 488          * FTP with NAT) adusting the TCP payload size when mangling IP
 489          * addresses and/or port numbers in the text-based control connection.
 490          */
 491         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 492             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 493                 return NF_DROP;
 494         return NF_ACCEPT;
 495 }
 496
 497 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 498  * value if 'skb' is freed.
 499  */
 500 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 501                             u16 zone, struct sk_buff *skb)
 502 {
 503         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 504         int err;
 505
 506         if (key->eth.type == htons(ETH_P_IP)) {
 507                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 508
 509                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 510                 err = ip_defrag(net, skb, user);
 511                 if (err)
 512                         return err;
 513
 514                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 515 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 516         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 517                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 518
 519                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 520                 err = nf_ct_frag6_gather(net, skb, user);
 521                 if (err) {
 522                         if (err != -EINPROGRESS)
 523                                 kfree_skb(skb);
 524                         return err;
 525                 }
 526
 527                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 528                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 529 #endif
 530         } else {
 531                 kfree_skb(skb);
 532                 return -EPFNOSUPPORT;
 533         }
 534
 535         key->ip.frag = OVS_FRAG_TYPE_NONE;
 536         skb_clear_hash(skb);
 537         skb->ignore_df = 1;
 538         *OVS_CB(skb) = ovs_cb;
 539
 540         return 0;
 541 }
 542
 543 static struct nf_conntrack_expect *
 544 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 545                    u16 proto, const struct sk_buff *skb)
 546 {
 547         struct nf_conntrack_tuple tuple;
 548         struct nf_conntrack_expect *exp;
 549
 550         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 551                 return NULL;
 552
 553         exp = __nf_ct_expect_find(net, zone, &tuple);
 554         if (exp) {
 555                 struct nf_conntrack_tuple_hash *h;
 556
 557                 /* Delete existing conntrack entry, if it clashes with the
 558                  * expectation.  This can happen since conntrack ALGs do not
 559                  * check for clashes between (new) expectations and existing
 560                  * conntrack entries.  nf_conntrack_in() will check the
 561                  * expectations only if a conntrack entry can not be found,
 562                  * which can lead to OVS finding the expectation (here) in the
 563                  * init direction, but which will not be removed by the
 564                  * nf_conntrack_in() call, if a matching conntrack entry is
 565                  * found instead.  In this case all init direction packets
 566                  * would be reported as new related packets, while reply
 567                  * direction packets would be reported as un-related
 568                  * established packets.
 569                  */
 570                 h = nf_conntrack_find_get(net, zone, &tuple);
 571                 if (h) {
 572                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 573
 574                         nf_ct_delete(ct, 0, 0);
 575                         nf_conntrack_put(&ct->ct_general);
 576                 }
 577         }
 578
 579         return exp;
 580 }
 581
 582 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 583 static enum ip_conntrack_info
 584 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 585 {
 586         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 587
 588         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 589                 return IP_CT_ESTABLISHED_REPLY;
 590         /* Once we've had two way comms, always ESTABLISHED. */
 591         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 592                 return IP_CT_ESTABLISHED;
 593         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 594                 return IP_CT_RELATED;
 595         return IP_CT_NEW;
 596 }
 597
 598 /* Find an existing connection which this packet belongs to without
 599  * re-attributing statistics or modifying the connection state.  This allows an
 600  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 601  *
 602  * Must be called with rcu_read_lock.
 603  *
 604  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 605  * if there is no existing entry.
 606  */
 607 static struct nf_conn *
 608 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 609                      u8 l3num, struct sk_buff *skb, bool natted)
 610 {
 611         struct nf_conntrack_tuple tuple;
 612         struct nf_conntrack_tuple_hash *h;
 613         struct nf_conn *ct;
 614
 615         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 616                                net, &tuple)) {
 617                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 618                 return NULL;
 619         }
 620
 621         /* Must invert the tuple if skb has been transformed by NAT. */
 622         if (natted) {
 623                 struct nf_conntrack_tuple inverse;
 624
 625                 if (!nf_ct_invert_tuplepr(&inverse, &tuple)) {
 626                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 627                         return NULL;
 628                 }
 629                 tuple = inverse;
 630         }
 631
 632         /* look for tuple match */
 633         h = nf_conntrack_find_get(net, zone, &tuple);
 634         if (!h)
 635                 return NULL;   /* Not found. */
 636
 637         ct = nf_ct_tuplehash_to_ctrack(h);
 638
 639         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 640          * select the other tuplehash to get the right 'ctinfo' bits for this
 641          * packet.
 642          */
 643         if (natted)
 644                 h = &ct->tuplehash[!h->tuple.dst.dir];
 645
 646         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 647         return ct;
 648 }
 649
 650 static
 651 struct nf_conn *ovs_ct_executed(struct net *net,
 652                                 const struct sw_flow_key *key,
 653                                 const struct ovs_conntrack_info *info,
 654                                 struct sk_buff *skb,
 655                                 bool *ct_executed)
 656 {
 657         struct nf_conn *ct = NULL;
 658
 659         /* If no ct, check if we have evidence that an existing conntrack entry
 660          * might be found for this skb.  This happens when we lose a skb->_nfct
 661          * due to an upcall, or if the direction is being forced.  If the
 662          * connection was not confirmed, it is not cached and needs to be run
 663          * through conntrack again.
 664          */
 665         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 666                        !(key->ct_state & OVS_CS_F_INVALID) &&
 667                        (key->ct_zone == info->zone.id);
 668
 669         if (*ct_executed || (!key->ct_state && info->force)) {
 670                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 671                                           !!(key->ct_state &
 672                                           OVS_CS_F_NAT_MASK));
 673         }
 674
 675         return ct;
 676 }
 677
 678 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 679 static bool skb_nfct_cached(struct net *net,
 680                             const struct sw_flow_key *key,
 681                             const struct ovs_conntrack_info *info,
 682                             struct sk_buff *skb)
 683 {
 684         enum ip_conntrack_info ctinfo;
 685         struct nf_conn *ct;
 686         bool ct_executed = true;
 687
 688         ct = nf_ct_get(skb, &ctinfo);
 689         if (!ct)
 690                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 691
 692         if (ct)
 693                 nf_ct_get(skb, &ctinfo);
 694         else
 695                 return false;
 696
 697         if (!net_eq(net, read_pnet(&ct->ct_net)))
 698                 return false;
 699         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 700                 return false;
 701         if (info->helper) {
 702                 struct nf_conn_help *help;
 703
 704                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 705                 if (help && rcu_access_pointer(help->helper) != info->helper)
 706                         return false;
 707         }
 708         /* Force conntrack entry direction to the current packet? */
 709         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 710                 /* Delete the conntrack entry if confirmed, else just release
 711                  * the reference.
 712                  */
 713                 if (nf_ct_is_confirmed(ct))
 714                         nf_ct_delete(ct, 0, 0);
 715
 716                 nf_conntrack_put(&ct->ct_general);
 717                 nf_ct_set(skb, NULL, 0);
 718                 return false;
 719         }
 720
 721         return ct_executed;
 722 }
 723
 724 #ifdef CONFIG_NF_NAT_NEEDED
 725 /* Modelled after nf_nat_ipv[46]_fn().
 726  * range is only used for new, uninitialized NAT state.
 727  * Returns either NF_ACCEPT or NF_DROP.
 728  */
 729 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 730                               enum ip_conntrack_info ctinfo,
 731                               const struct nf_nat_range2 *range,
 732                               enum nf_nat_manip_type maniptype)
 733 {
 734         int hooknum, nh_off, err = NF_ACCEPT;
 735
 736         nh_off = skb_network_offset(skb);
 737         skb_pull_rcsum(skb, nh_off);
 738
 739         /* See HOOK2MANIP(). */
 740         if (maniptype == NF_NAT_MANIP_SRC)
 741                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 742         else
 743                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 744
 745         switch (ctinfo) {
 746         case IP_CT_RELATED:
 747         case IP_CT_RELATED_REPLY:
 748                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 749                     skb->protocol == htons(ETH_P_IP) &&
 750                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 751                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 752                                                            hooknum))
 753                                 err = NF_DROP;
 754                         goto push;
 755                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 756                            skb->protocol == htons(ETH_P_IPV6)) {
 757                         __be16 frag_off;
 758                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 759                         int hdrlen = ipv6_skip_exthdr(skb,
 760                                                       sizeof(struct ipv6hdr),
 761                                                       &nexthdr, &frag_off);
 762
 763                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 764                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 765                                                                      ctinfo,
 766                                                                      hooknum,
 767                                                                      hdrlen))
 768                                         err = NF_DROP;
 769                                 goto push;
 770                         }
 771                 }
 772                 /* Non-ICMP, fall thru to initialize if needed. */
 773                 /* fall through */
 774         case IP_CT_NEW:
 775                 /* Seen it before?  This can happen for loopback, retrans,
 776                  * or local packets.
 777                  */
 778                 if (!nf_nat_initialized(ct, maniptype)) {
 779                         /* Initialize according to the NAT action. */
 780                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 781                                 /* Action is set up to establish a new
 782                                  * mapping.
 783                                  */
 784                                 ? nf_nat_setup_info(ct, range, maniptype)
 785                                 : nf_nat_alloc_null_binding(ct, hooknum);
 786                         if (err != NF_ACCEPT)
 787                                 goto push;
 788                 }
 789                 break;
 790
 791         case IP_CT_ESTABLISHED:
 792         case IP_CT_ESTABLISHED_REPLY:
 793                 break;
 794
 795         default:
 796                 err = NF_DROP;
 797                 goto push;
 798         }
 799
 800         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 801 push:
 802         skb_push(skb, nh_off);
 803         skb_postpush_rcsum(skb, skb->data, nh_off);
 804
 805         return err;
 806 }
 807
 808 static void ovs_nat_update_key(struct sw_flow_key *key,
 809                                const struct sk_buff *skb,
 810                                enum nf_nat_manip_type maniptype)
 811 {
 812         if (maniptype == NF_NAT_MANIP_SRC) {
 813                 __be16 src;
 814
 815                 key->ct_state |= OVS_CS_F_SRC_NAT;
 816                 if (key->eth.type == htons(ETH_P_IP))
 817                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 818                 else if (key->eth.type == htons(ETH_P_IPV6))
 819                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 820                                sizeof(key->ipv6.addr.src));
 821                 else
 822                         return;
 823
 824                 if (key->ip.proto == IPPROTO_UDP)
 825                         src = udp_hdr(skb)->source;
 826                 else if (key->ip.proto == IPPROTO_TCP)
 827                         src = tcp_hdr(skb)->source;
 828                 else if (key->ip.proto == IPPROTO_SCTP)
 829                         src = sctp_hdr(skb)->source;
 830                 else
 831                         return;
 832
 833                 key->tp.src = src;
 834         } else {
 835                 __be16 dst;
 836
 837                 key->ct_state |= OVS_CS_F_DST_NAT;
 838                 if (key->eth.type == htons(ETH_P_IP))
 839                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 840                 else if (key->eth.type == htons(ETH_P_IPV6))
 841                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 842                                sizeof(key->ipv6.addr.dst));
 843                 else
 844                         return;
 845
 846                 if (key->ip.proto == IPPROTO_UDP)
 847                         dst = udp_hdr(skb)->dest;
 848                 else if (key->ip.proto == IPPROTO_TCP)
 849                         dst = tcp_hdr(skb)->dest;
 850                 else if (key->ip.proto == IPPROTO_SCTP)
 851                         dst = sctp_hdr(skb)->dest;
 852                 else
 853                         return;
 854
 855                 key->tp.dst = dst;
 856         }
 857 }
 858
 859 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 860 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 861                       const struct ovs_conntrack_info *info,
 862                       struct sk_buff *skb, struct nf_conn *ct,
 863                       enum ip_conntrack_info ctinfo)
 864 {
 865         enum nf_nat_manip_type maniptype;
 866         int err;
 867
 868         /* Add NAT extension if not confirmed yet. */
 869         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 870                 return NF_ACCEPT;   /* Can't NAT. */
 871
 872         /* Determine NAT type.
 873          * Check if the NAT type can be deduced from the tracked connection.
 874          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 875          * when committing.
 876          */
 877         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 878             ct->status & IPS_NAT_MASK &&
 879             (ctinfo != IP_CT_RELATED || info->commit)) {
 880                 /* NAT an established or related connection like before. */
 881                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 882                         /* This is the REPLY direction for a connection
 883                          * for which NAT was applied in the forward
 884                          * direction.  Do the reverse NAT.
 885                          */
 886                         maniptype = ct->status & IPS_SRC_NAT
 887                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 888                 else
 889                         maniptype = ct->status & IPS_SRC_NAT
 890                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 891         } else if (info->nat & OVS_CT_SRC_NAT) {
 892                 maniptype = NF_NAT_MANIP_SRC;
 893         } else if (info->nat & OVS_CT_DST_NAT) {
 894                 maniptype = NF_NAT_MANIP_DST;
 895         } else {
 896                 return NF_ACCEPT; /* Connection is not NATed. */
 897         }
 898         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 899
 900         /* Mark NAT done if successful and update the flow key. */
 901         if (err == NF_ACCEPT)
 902                 ovs_nat_update_key(key, skb, maniptype);
 903
 904         return err;
 905 }
 906 #else /* !CONFIG_NF_NAT_NEEDED */
 907 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 908                       const struct ovs_conntrack_info *info,
 909                       struct sk_buff *skb, struct nf_conn *ct,
 910                       enum ip_conntrack_info ctinfo)
 911 {
 912         return NF_ACCEPT;
 913 }
 914 #endif
 915
 916 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 917  * not done already.  Update key with new CT state after passing the packet
 918  * through conntrack.
 919  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 920  * set to NULL and 0 will be returned.
 921  */
 922 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 923                            const struct ovs_conntrack_info *info,
 924                            struct sk_buff *skb)
 925 {
 926         /* If we are recirculating packets to match on conntrack fields and
 927          * committing with a separate conntrack action,  then we don't need to
 928          * actually run the packet through conntrack twice unless it's for a
 929          * different zone.
 930          */
 931         bool cached = skb_nfct_cached(net, key, info, skb);
 932         enum ip_conntrack_info ctinfo;
 933         struct nf_conn *ct;
 934
 935         if (!cached) {
 936                 struct nf_conn *tmpl = info->ct;
 937                 int err;
 938
 939                 /* Associate skb with specified zone. */
 940                 if (tmpl) {
 941                         if (skb_nfct(skb))
 942                                 nf_conntrack_put(skb_nfct(skb));
 943                         nf_conntrack_get(&tmpl->ct_general);
 944                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 945                 }
 946
 947                 err = nf_conntrack_in(net, info->family,
 948                                       NF_INET_PRE_ROUTING, skb);
 949                 if (err != NF_ACCEPT)
 950                         return -ENOENT;
 951
 952                 /* Clear CT state NAT flags to mark that we have not yet done
 953                  * NAT after the nf_conntrack_in() call.  We can actually clear
 954                  * the whole state, as it will be re-initialized below.
 955                  */
 956                 key->ct_state = 0;
 957
 958                 /* Update the key, but keep the NAT flags. */
 959                 ovs_ct_update_key(skb, info, key, true, true);
 960         }
 961
 962         ct = nf_ct_get(skb, &ctinfo);
 963         if (ct) {
 964                 /* Packets starting a new connection must be NATted before the
 965                  * helper, so that the helper knows about the NAT.  We enforce
 966                  * this by delaying both NAT and helper calls for unconfirmed
 967                  * connections until the committing CT action.  For later
 968                  * packets NAT and Helper may be called in either order.
 969                  *
 970                  * NAT will be done only if the CT action has NAT, and only
 971                  * once per packet (per zone), as guarded by the NAT bits in
 972                  * the key->ct_state.
 973                  */
 974                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 975                     (nf_ct_is_confirmed(ct) || info->commit) &&
 976                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 977                         return -EINVAL;
 978                 }
 979
 980                 /* Userspace may decide to perform a ct lookup without a helper
 981                  * specified followed by a (recirculate and) commit with one.
 982                  * Therefore, for unconfirmed connections which we will commit,
 983                  * we need to attach the helper here.
 984                  */
 985                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 986                     info->helper && !nfct_help(ct)) {
 987                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 988                                                             GFP_ATOMIC);
 989                         if (err)
 990                                 return err;
 991                 }
 992
 993                 /* Call the helper only if:
 994                  * - nf_conntrack_in() was executed above ("!cached") for a
 995                  *   confirmed connection, or
 996                  * - When committing an unconfirmed connection.
 997                  */
 998                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 999                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1000                         return -EINVAL;
1001                 }
1002         }
1003
1004         return 0;
1005 }
1006
1007 /* Lookup connection and read fields into key. */
1008 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1009                          const struct ovs_conntrack_info *info,
1010                          struct sk_buff *skb)
1011 {
1012         struct nf_conntrack_expect *exp;
1013
1014         /* If we pass an expected packet through nf_conntrack_in() the
1015          * expectation is typically removed, but the packet could still be
1016          * lost in upcall processing.  To prevent this from happening we
1017          * perform an explicit expectation lookup.  Expected connections are
1018          * always new, and will be passed through conntrack only when they are
1019          * committed, as it is OK to remove the expectation at that time.
1020          */
1021         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1022         if (exp) {
1023                 u8 state;
1024
1025                 /* NOTE: New connections are NATted and Helped only when
1026                  * committed, so we are not calling into NAT here.
1027                  */
1028                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1029                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1030         } else {
1031                 struct nf_conn *ct;
1032                 int err;
1033
1034                 err = __ovs_ct_lookup(net, key, info, skb);
1035                 if (err)
1036                         return err;
1037
1038                 ct = (struct nf_conn *)skb_nfct(skb);
1039                 if (ct)
1040                         nf_ct_deliver_cached_events(ct);
1041         }
1042
1043         return 0;
1044 }
1045
1046 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1047 {
1048         size_t i;
1049
1050         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1051                 if (labels->ct_labels_32[i])
1052                         return true;
1053
1054         return false;
1055 }
1056
1057 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1058 static struct hlist_head *ct_limit_hash_bucket(
1059         const struct ovs_ct_limit_info *info, u16 zone)
1060 {
1061         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1062 }
1063
1064 /* Call with ovs_mutex */
1065 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1066                          struct ovs_ct_limit *new_ct_limit)
1067 {
1068         struct ovs_ct_limit *ct_limit;
1069         struct hlist_head *head;
1070
1071         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1072         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1073                 if (ct_limit->zone == new_ct_limit->zone) {
1074                         hlist_replace_rcu(&ct_limit->hlist_node,
1075                                           &new_ct_limit->hlist_node);
1076                         kfree_rcu(ct_limit, rcu);
1077                         return;
1078                 }
1079         }
1080
1081         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1082 }
1083
1084 /* Call with ovs_mutex */
1085 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1086 {
1087         struct ovs_ct_limit *ct_limit;
1088         struct hlist_head *head;
1089         struct hlist_node *n;
1090
1091         head = ct_limit_hash_bucket(info, zone);
1092         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1093                 if (ct_limit->zone == zone) {
1094                         hlist_del_rcu(&ct_limit->hlist_node);
1095                         kfree_rcu(ct_limit, rcu);
1096                         return;
1097                 }
1098         }
1099 }
1100
1101 /* Call with RCU read lock */
1102 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1103 {
1104         struct ovs_ct_limit *ct_limit;
1105         struct hlist_head *head;
1106
1107         head = ct_limit_hash_bucket(info, zone);
1108         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1109                 if (ct_limit->zone == zone)
1110                         return ct_limit->limit;
1111         }
1112
1113         return info->default_limit;
1114 }
1115
1116 static int ovs_ct_check_limit(struct net *net,
1117                               const struct ovs_conntrack_info *info,
1118                               const struct nf_conntrack_tuple *tuple)
1119 {
1120         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1121         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1122         u32 per_zone_limit, connections;
1123         u32 conncount_key;
1124
1125         conncount_key = info->zone.id;
1126
1127         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1128         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1129                 return 0;
1130
1131         connections = nf_conncount_count(net, ct_limit_info->data,
1132                                          &conncount_key, tuple, &info->zone);
1133         if (connections > per_zone_limit)
1134                 return -ENOMEM;
1135
1136         return 0;
1137 }
1138 #endif
1139
1140 /* Lookup connection and confirm if unconfirmed. */
1141 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1142                          const struct ovs_conntrack_info *info,
1143                          struct sk_buff *skb)
1144 {
1145         enum ip_conntrack_info ctinfo;
1146         struct nf_conn *ct;
1147         int err;
1148
1149         err = __ovs_ct_lookup(net, key, info, skb);
1150         if (err)
1151                 return err;
1152
1153         /* The connection could be invalid, in which case this is a no-op.*/
1154         ct = nf_ct_get(skb, &ctinfo);
1155         if (!ct)
1156                 return 0;
1157
1158 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1159         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1160                 if (!nf_ct_is_confirmed(ct)) {
1161                         err = ovs_ct_check_limit(net, info,
1162                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1163                         if (err) {
1164                                 net_warn_ratelimited("openvswitch: zone: %u "
1165                                         "execeeds conntrack limit\n",
1166                                         info->zone.id);
1167                                 return err;
1168                         }
1169                 }
1170         }
1171 #endif
1172
1173         /* Set the conntrack event mask if given.  NEW and DELETE events have
1174          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1175          * typically would receive many kinds of updates.  Setting the event
1176          * mask allows those events to be filtered.  The set event mask will
1177          * remain in effect for the lifetime of the connection unless changed
1178          * by a further CT action with both the commit flag and the eventmask
1179          * option. */
1180         if (info->have_eventmask) {
1181                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1182
1183                 if (cache)
1184                         cache->ctmask = info->eventmask;
1185         }
1186
1187         /* Apply changes before confirming the connection so that the initial
1188          * conntrack NEW netlink event carries the values given in the CT
1189          * action.
1190          */
1191         if (info->mark.mask) {
1192                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1193                                       info->mark.mask);
1194                 if (err)
1195                         return err;
1196         }
1197         if (!nf_ct_is_confirmed(ct)) {
1198                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1199                                          &info->labels.mask);
1200                 if (err)
1201                         return err;
1202         } else if (labels_nonzero(&info->labels.mask)) {
1203                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1204                                         &info->labels.mask);
1205                 if (err)
1206                         return err;
1207         }
1208         /* This will take care of sending queued events even if the connection
1209          * is already confirmed.
1210          */
1211         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1212                 return -EINVAL;
1213
1214         return 0;
1215 }
1216
1217 /* Trim the skb to the length specified by the IP/IPv6 header,
1218  * removing any trailing lower-layer padding. This prepares the skb
1219  * for higher-layer processing that assumes skb->len excludes padding
1220  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1221  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1222  */
1223 static int ovs_skb_network_trim(struct sk_buff *skb)
1224 {
1225         unsigned int len;
1226         int err;
1227
1228         switch (skb->protocol) {
1229         case htons(ETH_P_IP):
1230                 len = ntohs(ip_hdr(skb)->tot_len);
1231                 break;
1232         case htons(ETH_P_IPV6):
1233                 len = sizeof(struct ipv6hdr)
1234                         + ntohs(ipv6_hdr(skb)->payload_len);
1235                 break;
1236         default:
1237                 len = skb->len;
1238         }
1239
1240         err = pskb_trim_rcsum(skb, len);
1241         if (err)
1242                 kfree_skb(skb);
1243
1244         return err;
1245 }
1246
1247 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1248  * value if 'skb' is freed.
1249  */
1250 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1251                    struct sw_flow_key *key,
1252                    const struct ovs_conntrack_info *info)
1253 {
1254         int nh_ofs;
1255         int err;
1256
1257         /* The conntrack module expects to be working at L3. */
1258         nh_ofs = skb_network_offset(skb);
1259         skb_pull_rcsum(skb, nh_ofs);
1260
1261         err = ovs_skb_network_trim(skb);
1262         if (err)
1263                 return err;
1264
1265         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1266                 err = handle_fragments(net, key, info->zone.id, skb);
1267                 if (err)
1268                         return err;
1269         }
1270
1271         if (info->commit)
1272                 err = ovs_ct_commit(net, key, info, skb);
1273         else
1274                 err = ovs_ct_lookup(net, key, info, skb);
1275
1276         skb_push(skb, nh_ofs);
1277         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1278         if (err)
1279                 kfree_skb(skb);
1280         return err;
1281 }
1282
1283 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1284 {
1285         if (skb_nfct(skb)) {
1286                 nf_conntrack_put(skb_nfct(skb));
1287                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1288                 ovs_ct_fill_key(skb, key);
1289         }
1290
1291         return 0;
1292 }
1293
1294 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1295                              const struct sw_flow_key *key, bool log)
1296 {
1297         struct nf_conntrack_helper *helper;
1298         struct nf_conn_help *help;
1299
1300         helper = nf_conntrack_helper_try_module_get(name, info->family,
1301                                                     key->ip.proto);
1302         if (!helper) {
1303                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1304                 return -EINVAL;
1305         }
1306
1307         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1308         if (!help) {
1309                 nf_conntrack_helper_put(helper);
1310                 return -ENOMEM;
1311         }
1312
1313         rcu_assign_pointer(help->helper, helper);
1314         info->helper = helper;
1315         return 0;
1316 }
1317
1318 #ifdef CONFIG_NF_NAT_NEEDED
1319 static int parse_nat(const struct nlattr *attr,
1320                      struct ovs_conntrack_info *info, bool log)
1321 {
1322         struct nlattr *a;
1323         int rem;
1324         bool have_ip_max = false;
1325         bool have_proto_max = false;
1326         bool ip_vers = (info->family == NFPROTO_IPV6);
1327
1328         nla_for_each_nested(a, attr, rem) {
1329                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1330                         [OVS_NAT_ATTR_SRC] = {0, 0},
1331                         [OVS_NAT_ATTR_DST] = {0, 0},
1332                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1333                                                  sizeof(struct in6_addr)},
1334                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1335                                                  sizeof(struct in6_addr)},
1336                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1337                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1338                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1339                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1340                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1341                 };
1342                 int type = nla_type(a);
1343
1344                 if (type > OVS_NAT_ATTR_MAX) {
1345                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1346                                   type, OVS_NAT_ATTR_MAX);
1347                         return -EINVAL;
1348                 }
1349
1350                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1351                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1352                                   type, nla_len(a),
1353                                   ovs_nat_attr_lens[type][ip_vers]);
1354                         return -EINVAL;
1355                 }
1356
1357                 switch (type) {
1358                 case OVS_NAT_ATTR_SRC:
1359                 case OVS_NAT_ATTR_DST:
1360                         if (info->nat) {
1361                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1362                                 return -ERANGE;
1363                         }
1364                         info->nat |= OVS_CT_NAT;
1365                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1366                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1367                         break;
1368
1369                 case OVS_NAT_ATTR_IP_MIN:
1370                         nla_memcpy(&info->range.min_addr, a,
1371                                    sizeof(info->range.min_addr));
1372                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1373                         break;
1374
1375                 case OVS_NAT_ATTR_IP_MAX:
1376                         have_ip_max = true;
1377                         nla_memcpy(&info->range.max_addr, a,
1378                                    sizeof(info->range.max_addr));
1379                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1380                         break;
1381
1382                 case OVS_NAT_ATTR_PROTO_MIN:
1383                         info->range.min_proto.all = htons(nla_get_u16(a));
1384                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1385                         break;
1386
1387                 case OVS_NAT_ATTR_PROTO_MAX:
1388                         have_proto_max = true;
1389                         info->range.max_proto.all = htons(nla_get_u16(a));
1390                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1391                         break;
1392
1393                 case OVS_NAT_ATTR_PERSISTENT:
1394                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1395                         break;
1396
1397                 case OVS_NAT_ATTR_PROTO_HASH:
1398                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1399                         break;
1400
1401                 case OVS_NAT_ATTR_PROTO_RANDOM:
1402                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1403                         break;
1404
1405                 default:
1406                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1407                         return -EINVAL;
1408                 }
1409         }
1410
1411         if (rem > 0) {
1412                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1413                 return -EINVAL;
1414         }
1415         if (!info->nat) {
1416                 /* Do not allow flags if no type is given. */
1417                 if (info->range.flags) {
1418                         OVS_NLERR(log,
1419                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1420                                   );
1421                         return -EINVAL;
1422                 }
1423                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1424         } else if (!info->commit) {
1425                 OVS_NLERR(log,
1426                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1427                           );
1428                 return -EINVAL;
1429         }
1430         /* Allow missing IP_MAX. */
1431         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1432                 memcpy(&info->range.max_addr, &info->range.min_addr,
1433                        sizeof(info->range.max_addr));
1434         }
1435         /* Allow missing PROTO_MAX. */
1436         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1437             !have_proto_max) {
1438                 info->range.max_proto.all = info->range.min_proto.all;
1439         }
1440         return 0;
1441 }
1442 #endif
1443
1444 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1445         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1446         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1447         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1448                                     .maxlen = sizeof(u16) },
1449         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1450                                     .maxlen = sizeof(struct md_mark) },
1451         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1452                                     .maxlen = sizeof(struct md_labels) },
1453         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1454                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1455 #ifdef CONFIG_NF_NAT_NEEDED
1456         /* NAT length is checked when parsing the nested attributes. */
1457         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1458 #endif
1459         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1460                                     .maxlen = sizeof(u32) },
1461 };
1462
1463 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1464                     const char **helper, bool log)
1465 {
1466         struct nlattr *a;
1467         int rem;
1468
1469         nla_for_each_nested(a, attr, rem) {
1470                 int type = nla_type(a);
1471                 int maxlen;
1472                 int minlen;
1473
1474                 if (type > OVS_CT_ATTR_MAX) {
1475                         OVS_NLERR(log,
1476                                   "Unknown conntrack attr (type=%d, max=%d)",
1477                                   type, OVS_CT_ATTR_MAX);
1478                         return -EINVAL;
1479                 }
1480
1481                 maxlen = ovs_ct_attr_lens[type].maxlen;
1482                 minlen = ovs_ct_attr_lens[type].minlen;
1483                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1484                         OVS_NLERR(log,
1485                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1486                                   type, nla_len(a), maxlen);
1487                         return -EINVAL;
1488                 }
1489
1490                 switch (type) {
1491                 case OVS_CT_ATTR_FORCE_COMMIT:
1492                         info->force = true;
1493                         /* fall through. */
1494                 case OVS_CT_ATTR_COMMIT:
1495                         info->commit = true;
1496                         break;
1497 #ifdef CONFIG_NF_CONNTRACK_ZONES
1498                 case OVS_CT_ATTR_ZONE:
1499                         info->zone.id = nla_get_u16(a);
1500                         break;
1501 #endif
1502 #ifdef CONFIG_NF_CONNTRACK_MARK
1503                 case OVS_CT_ATTR_MARK: {
1504                         struct md_mark *mark = nla_data(a);
1505
1506                         if (!mark->mask) {
1507                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1508                                 return -EINVAL;
1509                         }
1510                         info->mark = *mark;
1511                         break;
1512                 }
1513 #endif
1514 #ifdef CONFIG_NF_CONNTRACK_LABELS
1515                 case OVS_CT_ATTR_LABELS: {
1516                         struct md_labels *labels = nla_data(a);
1517
1518                         if (!labels_nonzero(&labels->mask)) {
1519                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1520                                 return -EINVAL;
1521                         }
1522                         info->labels = *labels;
1523                         break;
1524                 }
1525 #endif
1526                 case OVS_CT_ATTR_HELPER:
1527                         *helper = nla_data(a);
1528                         if (!memchr(*helper, '\0', nla_len(a))) {
1529                                 OVS_NLERR(log, "Invalid conntrack helper");
1530                                 return -EINVAL;
1531                         }
1532                         break;
1533 #ifdef CONFIG_NF_NAT_NEEDED
1534                 case OVS_CT_ATTR_NAT: {
1535                         int err = parse_nat(a, info, log);
1536
1537                         if (err)
1538                                 return err;
1539                         break;
1540                 }
1541 #endif
1542                 case OVS_CT_ATTR_EVENTMASK:
1543                         info->have_eventmask = true;
1544                         info->eventmask = nla_get_u32(a);
1545                         break;
1546
1547                 default:
1548                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1549                                   type);
1550                         return -EINVAL;
1551                 }
1552         }
1553
1554 #ifdef CONFIG_NF_CONNTRACK_MARK
1555         if (!info->commit && info->mark.mask) {
1556                 OVS_NLERR(log,
1557                           "Setting conntrack mark requires 'commit' flag.");
1558                 return -EINVAL;
1559         }
1560 #endif
1561 #ifdef CONFIG_NF_CONNTRACK_LABELS
1562         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1563                 OVS_NLERR(log,
1564                           "Setting conntrack labels requires 'commit' flag.");
1565                 return -EINVAL;
1566         }
1567 #endif
1568         if (rem > 0) {
1569                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1570                 return -EINVAL;
1571         }
1572
1573         return 0;
1574 }
1575
1576 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1577 {
1578         if (attr == OVS_KEY_ATTR_CT_STATE)
1579                 return true;
1580         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1581             attr == OVS_KEY_ATTR_CT_ZONE)
1582                 return true;
1583         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1584             attr == OVS_KEY_ATTR_CT_MARK)
1585                 return true;
1586         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1587             attr == OVS_KEY_ATTR_CT_LABELS) {
1588                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1589
1590                 return ovs_net->xt_label;
1591         }
1592
1593         return false;
1594 }
1595
1596 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1597                        const struct sw_flow_key *key,
1598                        struct sw_flow_actions **sfa,  bool log)
1599 {
1600         struct ovs_conntrack_info ct_info;
1601         const char *helper = NULL;
1602         u16 family;
1603         int err;
1604
1605         family = key_to_nfproto(key);
1606         if (family == NFPROTO_UNSPEC) {
1607                 OVS_NLERR(log, "ct family unspecified");
1608                 return -EINVAL;
1609         }
1610
1611         memset(&ct_info, 0, sizeof(ct_info));
1612         ct_info.family = family;
1613
1614         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1615                         NF_CT_DEFAULT_ZONE_DIR, 0);
1616
1617         err = parse_ct(attr, &ct_info, &helper, log);
1618         if (err)
1619                 return err;
1620
1621         /* Set up template for tracking connections in specific zones. */
1622         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1623         if (!ct_info.ct) {
1624                 OVS_NLERR(log, "Failed to allocate conntrack template");
1625                 return -ENOMEM;
1626         }
1627
1628         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1629         nf_conntrack_get(&ct_info.ct->ct_general);
1630
1631         if (helper) {
1632                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1633                 if (err)
1634                         goto err_free_ct;
1635         }
1636
1637         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1638                                  sizeof(ct_info), log);
1639         if (err)
1640                 goto err_free_ct;
1641
1642         return 0;
1643 err_free_ct:
1644         __ovs_ct_free_action(&ct_info);
1645         return err;
1646 }
1647
1648 #ifdef CONFIG_NF_NAT_NEEDED
1649 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1650                                struct sk_buff *skb)
1651 {
1652         struct nlattr *start;
1653
1654         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1655         if (!start)
1656                 return false;
1657
1658         if (info->nat & OVS_CT_SRC_NAT) {
1659                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1660                         return false;
1661         } else if (info->nat & OVS_CT_DST_NAT) {
1662                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1663                         return false;
1664         } else {
1665                 goto out;
1666         }
1667
1668         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1669                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1670                     info->family == NFPROTO_IPV4) {
1671                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1672                                             info->range.min_addr.ip) ||
1673                             (info->range.max_addr.ip
1674                              != info->range.min_addr.ip &&
1675                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1676                                               info->range.max_addr.ip))))
1677                                 return false;
1678                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1679                            info->family == NFPROTO_IPV6) {
1680                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1681                                              &info->range.min_addr.in6) ||
1682                             (memcmp(&info->range.max_addr.in6,
1683                                     &info->range.min_addr.in6,
1684                                     sizeof(info->range.max_addr.in6)) &&
1685                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1686                                                &info->range.max_addr.in6))))
1687                                 return false;
1688                 } else {
1689                         return false;
1690                 }
1691         }
1692         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1693             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1694                          ntohs(info->range.min_proto.all)) ||
1695              (info->range.max_proto.all != info->range.min_proto.all &&
1696               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1697                           ntohs(info->range.max_proto.all)))))
1698                 return false;
1699
1700         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1701             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1702                 return false;
1703         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1704             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1705                 return false;
1706         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1707             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1708                 return false;
1709 out:
1710         nla_nest_end(skb, start);
1711
1712         return true;
1713 }
1714 #endif
1715
1716 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1717                           struct sk_buff *skb)
1718 {
1719         struct nlattr *start;
1720
1721         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1722         if (!start)
1723                 return -EMSGSIZE;
1724
1725         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1726                                             ? OVS_CT_ATTR_FORCE_COMMIT
1727                                             : OVS_CT_ATTR_COMMIT))
1728                 return -EMSGSIZE;
1729         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1730             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1731                 return -EMSGSIZE;
1732         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1733             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1734                     &ct_info->mark))
1735                 return -EMSGSIZE;
1736         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1737             labels_nonzero(&ct_info->labels.mask) &&
1738             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1739                     &ct_info->labels))
1740                 return -EMSGSIZE;
1741         if (ct_info->helper) {
1742                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1743                                    ct_info->helper->name))
1744                         return -EMSGSIZE;
1745         }
1746         if (ct_info->have_eventmask &&
1747             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1748                 return -EMSGSIZE;
1749
1750 #ifdef CONFIG_NF_NAT_NEEDED
1751         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1752                 return -EMSGSIZE;
1753 #endif
1754         nla_nest_end(skb, start);
1755
1756         return 0;
1757 }
1758
1759 void ovs_ct_free_action(const struct nlattr *a)
1760 {
1761         struct ovs_conntrack_info *ct_info = nla_data(a);
1762
1763         __ovs_ct_free_action(ct_info);
1764 }
1765
1766 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1767 {
1768         if (ct_info->helper)
1769                 nf_conntrack_helper_put(ct_info->helper);
1770         if (ct_info->ct)
1771                 nf_ct_tmpl_free(ct_info->ct);
1772 }
1773
1774 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1775 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1776 {
1777         int i, err;
1778
1779         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1780                                          GFP_KERNEL);
1781         if (!ovs_net->ct_limit_info)
1782                 return -ENOMEM;
1783
1784         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1785         ovs_net->ct_limit_info->limits =
1786                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1787                               GFP_KERNEL);
1788         if (!ovs_net->ct_limit_info->limits) {
1789                 kfree(ovs_net->ct_limit_info);
1790                 return -ENOMEM;
1791         }
1792
1793         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1794                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1795
1796         ovs_net->ct_limit_info->data =
1797                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1798
1799         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1800                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1801                 kfree(ovs_net->ct_limit_info->limits);
1802                 kfree(ovs_net->ct_limit_info);
1803                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1804                 return err;
1805         }
1806         return 0;
1807 }
1808
1809 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1810 {
1811         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1812         int i;
1813
1814         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1815         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1816                 struct hlist_head *head = &info->limits[i];
1817                 struct ovs_ct_limit *ct_limit;
1818
1819                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1820                         kfree_rcu(ct_limit, rcu);
1821         }
1822         kfree(ovs_net->ct_limit_info->limits);
1823         kfree(ovs_net->ct_limit_info);
1824 }
1825
1826 static struct sk_buff *
1827 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1828                              struct ovs_header **ovs_reply_header)
1829 {
1830         struct ovs_header *ovs_header = info->userhdr;
1831         struct sk_buff *skb;
1832
1833         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1834         if (!skb)
1835                 return ERR_PTR(-ENOMEM);
1836
1837         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1838                                         info->snd_seq,
1839                                         &dp_ct_limit_genl_family, 0, cmd);
1840
1841         if (!*ovs_reply_header) {
1842                 nlmsg_free(skb);
1843                 return ERR_PTR(-EMSGSIZE);
1844         }
1845         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1846
1847         return skb;
1848 }
1849
1850 static bool check_zone_id(int zone_id, u16 *pzone)
1851 {
1852         if (zone_id >= 0 && zone_id <= 65535) {
1853                 *pzone = (u16)zone_id;
1854                 return true;
1855         }
1856         return false;
1857 }
1858
1859 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1860                                        struct ovs_ct_limit_info *info)
1861 {
1862         struct ovs_zone_limit *zone_limit;
1863         int rem;
1864         u16 zone;
1865
1866         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1867         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1868
1869         while (rem >= sizeof(*zone_limit)) {
1870                 if (unlikely(zone_limit->zone_id ==
1871                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1872                         ovs_lock();
1873                         info->default_limit = zone_limit->limit;
1874                         ovs_unlock();
1875                 } else if (unlikely(!check_zone_id(
1876                                 zone_limit->zone_id, &zone))) {
1877                         OVS_NLERR(true, "zone id is out of range");
1878                 } else {
1879                         struct ovs_ct_limit *ct_limit;
1880
1881                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1882                         if (!ct_limit)
1883                                 return -ENOMEM;
1884
1885                         ct_limit->zone = zone;
1886                         ct_limit->limit = zone_limit->limit;
1887
1888                         ovs_lock();
1889                         ct_limit_set(info, ct_limit);
1890                         ovs_unlock();
1891                 }
1892                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1893                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1894                                 NLA_ALIGN(sizeof(*zone_limit)));
1895         }
1896
1897         if (rem)
1898                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1899
1900         return 0;
1901 }
1902
1903 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1904                                        struct ovs_ct_limit_info *info)
1905 {
1906         struct ovs_zone_limit *zone_limit;
1907         int rem;
1908         u16 zone;
1909
1910         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1911         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1912
1913         while (rem >= sizeof(*zone_limit)) {
1914                 if (unlikely(zone_limit->zone_id ==
1915                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1916                         ovs_lock();
1917                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1918                         ovs_unlock();
1919                 } else if (unlikely(!check_zone_id(
1920                                 zone_limit->zone_id, &zone))) {
1921                         OVS_NLERR(true, "zone id is out of range");
1922                 } else {
1923                         ovs_lock();
1924                         ct_limit_del(info, zone);
1925                         ovs_unlock();
1926                 }
1927                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1928                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1929                                 NLA_ALIGN(sizeof(*zone_limit)));
1930         }
1931
1932         if (rem)
1933                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1934
1935         return 0;
1936 }
1937
1938 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1939                                           struct sk_buff *reply)
1940 {
1941         struct ovs_zone_limit zone_limit;
1942         int err;
1943
1944         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1945         zone_limit.limit = info->default_limit;
1946         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1947         if (err)
1948                 return err;
1949
1950         return 0;
1951 }
1952
1953 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1954                                          struct nf_conncount_data *data,
1955                                          u16 zone_id, u32 limit,
1956                                          struct sk_buff *reply)
1957 {
1958         struct nf_conntrack_zone ct_zone;
1959         struct ovs_zone_limit zone_limit;
1960         u32 conncount_key = zone_id;
1961
1962         zone_limit.zone_id = zone_id;
1963         zone_limit.limit = limit;
1964         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1965
1966         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1967                                               &ct_zone);
1968         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1969 }
1970
1971 static int ovs_ct_limit_get_zone_limit(struct net *net,
1972                                        struct nlattr *nla_zone_limit,
1973                                        struct ovs_ct_limit_info *info,
1974                                        struct sk_buff *reply)
1975 {
1976         struct ovs_zone_limit *zone_limit;
1977         int rem, err;
1978         u32 limit;
1979         u16 zone;
1980
1981         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1982         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1983
1984         while (rem >= sizeof(*zone_limit)) {
1985                 if (unlikely(zone_limit->zone_id ==
1986                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1987                         err = ovs_ct_limit_get_default_limit(info, reply);
1988                         if (err)
1989                                 return err;
1990                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
1991                                                         &zone))) {
1992                         OVS_NLERR(true, "zone id is out of range");
1993                 } else {
1994                         rcu_read_lock();
1995                         limit = ct_limit_get(info, zone);
1996                         rcu_read_unlock();
1997
1998                         err = __ovs_ct_limit_get_zone_limit(
1999                                 net, info->data, zone, limit, reply);
2000                         if (err)
2001                                 return err;
2002                 }
2003                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2004                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2005                                 NLA_ALIGN(sizeof(*zone_limit)));
2006         }
2007
2008         if (rem)
2009                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2010
2011         return 0;
2012 }
2013
2014 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2015                                            struct ovs_ct_limit_info *info,
2016                                            struct sk_buff *reply)
2017 {
2018         struct ovs_ct_limit *ct_limit;
2019         struct hlist_head *head;
2020         int i, err = 0;
2021
2022         err = ovs_ct_limit_get_default_limit(info, reply);
2023         if (err)
2024                 return err;
2025
2026         rcu_read_lock();
2027         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2028                 head = &info->limits[i];
2029                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2030                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2031                                 ct_limit->zone, ct_limit->limit, reply);
2032                         if (err)
2033                                 goto exit_err;
2034                 }
2035         }
2036
2037 exit_err:
2038         rcu_read_unlock();
2039         return err;
2040 }
2041
2042 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2043 {
2044         struct nlattr **a = info->attrs;
2045         struct sk_buff *reply;
2046         struct ovs_header *ovs_reply_header;
2047         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2048         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2049         int err;
2050
2051         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2052                                              &ovs_reply_header);
2053         if (IS_ERR(reply))
2054                 return PTR_ERR(reply);
2055
2056         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2057                 err = -EINVAL;
2058                 goto exit_err;
2059         }
2060
2061         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2062                                           ct_limit_info);
2063         if (err)
2064                 goto exit_err;
2065
2066         static_branch_enable(&ovs_ct_limit_enabled);
2067
2068         genlmsg_end(reply, ovs_reply_header);
2069         return genlmsg_reply(reply, info);
2070
2071 exit_err:
2072         nlmsg_free(reply);
2073         return err;
2074 }
2075
2076 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2077 {
2078         struct nlattr **a = info->attrs;
2079         struct sk_buff *reply;
2080         struct ovs_header *ovs_reply_header;
2081         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2082         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2083         int err;
2084
2085         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2086                                              &ovs_reply_header);
2087         if (IS_ERR(reply))
2088                 return PTR_ERR(reply);
2089
2090         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2091                 err = -EINVAL;
2092                 goto exit_err;
2093         }
2094
2095         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2096                                           ct_limit_info);
2097         if (err)
2098                 goto exit_err;
2099
2100         genlmsg_end(reply, ovs_reply_header);
2101         return genlmsg_reply(reply, info);
2102
2103 exit_err:
2104         nlmsg_free(reply);
2105         return err;
2106 }
2107
2108 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2109 {
2110         struct nlattr **a = info->attrs;
2111         struct nlattr *nla_reply;
2112         struct sk_buff *reply;
2113         struct ovs_header *ovs_reply_header;
2114         struct net *net = sock_net(skb->sk);
2115         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2116         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2117         int err;
2118
2119         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2120                                              &ovs_reply_header);
2121         if (IS_ERR(reply))
2122                 return PTR_ERR(reply);
2123
2124         nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2125
2126         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2127                 err = ovs_ct_limit_get_zone_limit(
2128                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2129                         reply);
2130                 if (err)
2131                         goto exit_err;
2132         } else {
2133                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2134                                                       reply);
2135                 if (err)
2136                         goto exit_err;
2137         }
2138
2139         nla_nest_end(reply, nla_reply);
2140         genlmsg_end(reply, ovs_reply_header);
2141         return genlmsg_reply(reply, info);
2142
2143 exit_err:
2144         nlmsg_free(reply);
2145         return err;
2146 }
2147
2148 static struct genl_ops ct_limit_genl_ops[] = {
2149         { .cmd = OVS_CT_LIMIT_CMD_SET,
2150                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2151                                            * privilege. */
2152                 .policy = ct_limit_policy,
2153                 .doit = ovs_ct_limit_cmd_set,
2154         },
2155         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2156                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2157                                            * privilege. */
2158                 .policy = ct_limit_policy,
2159                 .doit = ovs_ct_limit_cmd_del,
2160         },
2161         { .cmd = OVS_CT_LIMIT_CMD_GET,
2162                 .flags = 0,               /* OK for unprivileged users. */
2163                 .policy = ct_limit_policy,
2164                 .doit = ovs_ct_limit_cmd_get,
2165         },
2166 };
2167
2168 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2169         .name = OVS_CT_LIMIT_MCGROUP,
2170 };
2171
2172 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2173         .hdrsize = sizeof(struct ovs_header),
2174         .name = OVS_CT_LIMIT_FAMILY,
2175         .version = OVS_CT_LIMIT_VERSION,
2176         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2177         .netnsok = true,
2178         .parallel_ops = true,
2179         .ops = ct_limit_genl_ops,
2180         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2181         .mcgrps = &ovs_ct_limit_multicast_group,
2182         .n_mcgrps = 1,
2183         .module = THIS_MODULE,
2184 };
2185 #endif
2186
2187 int ovs_ct_init(struct net *net)
2188 {
2189         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2190         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2191
2192         if (nf_connlabels_get(net, n_bits - 1)) {
2193                 ovs_net->xt_label = false;
2194                 OVS_NLERR(true, "Failed to set connlabel length");
2195         } else {
2196                 ovs_net->xt_label = true;
2197         }
2198
2199 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2200         return ovs_ct_limit_init(net, ovs_net);
2201 #else
2202         return 0;
2203 #endif
2204 }
2205
2206 void ovs_ct_exit(struct net *net)
2207 {
2208         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2209
2210 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2211         ovs_ct_limit_exit(net, ovs_net);
2212 #endif
2213
2214         if (ovs_net->xt_label)
2215                 nf_connlabels_put(net);
2216 }