net/openvswitch/conntrack.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2015 Nicira, Inc.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/openvswitch.h>
   8 #include <linux/tcp.h>
   9 #include <linux/udp.h>
  10 #include <linux/sctp.h>
  11 #include <linux/static_key.h>
  12 #include <net/ip.h>
  13 #include <net/genetlink.h>
  14 #include <net/netfilter/nf_conntrack_core.h>
  15 #include <net/netfilter/nf_conntrack_count.h>
  16 #include <net/netfilter/nf_conntrack_helper.h>
  17 #include <net/netfilter/nf_conntrack_labels.h>
  18 #include <net/netfilter/nf_conntrack_seqadj.h>
  19 #include <net/netfilter/nf_conntrack_timeout.h>
  20 #include <net/netfilter/nf_conntrack_zones.h>
  21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  22 #include <net/ipv6_frag.h>
  23
  24 #if IS_ENABLED(CONFIG_NF_NAT)
  25 #include <net/netfilter/nf_nat.h>
  26 #endif
  27
  28 #include "datapath.h"
  29 #include "conntrack.h"
  30 #include "flow.h"
  31 #include "flow_netlink.h"
  32
  33 struct ovs_ct_len_tbl {
  34         int maxlen;
  35         int minlen;
  36 };
  37
  38 /* Metadata mark for masked write to conntrack mark */
  39 struct md_mark {
  40         u32 value;
  41         u32 mask;
  42 };
  43
  44 /* Metadata label for masked write to conntrack label. */
  45 struct md_labels {
  46         struct ovs_key_ct_labels value;
  47         struct ovs_key_ct_labels mask;
  48 };
  49
  50 enum ovs_ct_nat {
  51         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  52         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  53         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  54 };
  55
  56 /* Conntrack action context for execution. */
  57 struct ovs_conntrack_info {
  58         struct nf_conntrack_helper *helper;
  59         struct nf_conntrack_zone zone;
  60         struct nf_conn *ct;
  61         u8 commit : 1;
  62         u8 nat : 3;                 /* enum ovs_ct_nat */
  63         u8 force : 1;
  64         u8 have_eventmask : 1;
  65         u16 family;
  66         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  67         struct md_mark mark;
  68         struct md_labels labels;
  69         char timeout[CTNL_TIMEOUT_NAME_MAX];
  70         struct nf_ct_timeout *nf_ct_timeout;
  71 #if IS_ENABLED(CONFIG_NF_NAT)
  72         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  73 #endif
  74 };
  75
  76 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  77 #define OVS_CT_LIMIT_UNLIMITED  0
  78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  79 #define CT_LIMIT_HASH_BUCKETS 512
  80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  81
  82 struct ovs_ct_limit {
  83         /* Elements in ovs_ct_limit_info->limits hash table */
  84         struct hlist_node hlist_node;
  85         struct rcu_head rcu;
  86         u16 zone;
  87         u32 limit;
  88 };
  89
  90 struct ovs_ct_limit_info {
  91         u32 default_limit;
  92         struct hlist_head *limits;
  93         struct nf_conncount_data *data;
  94 };
  95
  96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
  97         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
  98 };
  99 #endif
 100
 101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 102
 103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 104
 105 static u16 key_to_nfproto(const struct sw_flow_key *key)
 106 {
 107         switch (ntohs(key->eth.type)) {
 108         case ETH_P_IP:
 109                 return NFPROTO_IPV4;
 110         case ETH_P_IPV6:
 111                 return NFPROTO_IPV6;
 112         default:
 113                 return NFPROTO_UNSPEC;
 114         }
 115 }
 116
 117 /* Map SKB connection state into the values used by flow definition. */
 118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 119 {
 120         u8 ct_state = OVS_CS_F_TRACKED;
 121
 122         switch (ctinfo) {
 123         case IP_CT_ESTABLISHED_REPLY:
 124         case IP_CT_RELATED_REPLY:
 125                 ct_state |= OVS_CS_F_REPLY_DIR;
 126                 break;
 127         default:
 128                 break;
 129         }
 130
 131         switch (ctinfo) {
 132         case IP_CT_ESTABLISHED:
 133         case IP_CT_ESTABLISHED_REPLY:
 134                 ct_state |= OVS_CS_F_ESTABLISHED;
 135                 break;
 136         case IP_CT_RELATED:
 137         case IP_CT_RELATED_REPLY:
 138                 ct_state |= OVS_CS_F_RELATED;
 139                 break;
 140         case IP_CT_NEW:
 141                 ct_state |= OVS_CS_F_NEW;
 142                 break;
 143         default:
 144                 break;
 145         }
 146
 147         return ct_state;
 148 }
 149
 150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 151 {
 152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 153         return ct ? ct->mark : 0;
 154 #else
 155         return 0;
 156 #endif
 157 }
 158
 159 /* Guard against conntrack labels max size shrinking below 128 bits. */
 160 #if NF_CT_LABELS_MAX_SIZE < 16
 161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 162 #endif
 163
 164 static void ovs_ct_get_labels(const struct nf_conn *ct,
 165                               struct ovs_key_ct_labels *labels)
 166 {
 167         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 168
 169         if (cl)
 170                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 171         else
 172                 memset(labels, 0, OVS_CT_LABELS_LEN);
 173 }
 174
 175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 176                                         const struct nf_conntrack_tuple *orig,
 177                                         u8 icmp_proto)
 178 {
 179         key->ct_orig_proto = orig->dst.protonum;
 180         if (orig->dst.protonum == icmp_proto) {
 181                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 182                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 183         } else {
 184                 key->ct.orig_tp.src = orig->src.u.all;
 185                 key->ct.orig_tp.dst = orig->dst.u.all;
 186         }
 187 }
 188
 189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 190                                 const struct nf_conntrack_zone *zone,
 191                                 const struct nf_conn *ct)
 192 {
 193         key->ct_state = state;
 194         key->ct_zone = zone->id;
 195         key->ct.mark = ovs_ct_get_mark(ct);
 196         ovs_ct_get_labels(ct, &key->ct.labels);
 197
 198         if (ct) {
 199                 const struct nf_conntrack_tuple *orig;
 200
 201                 /* Use the master if we have one. */
 202                 if (ct->master)
 203                         ct = ct->master;
 204                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 205
 206                 /* IP version must match with the master connection. */
 207                 if (key->eth.type == htons(ETH_P_IP) &&
 208                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 209                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 210                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 211                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 212                         return;
 213                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 214                            !sw_flow_key_is_nd(key) &&
 215                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 216                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 217                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 218                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 219                         return;
 220                 }
 221         }
 222         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 223          * original direction key fields.
 224          */
 225         key->ct_orig_proto = 0;
 226 }
 227
 228 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 229  * previously sent the packet to conntrack via the ct action.  If
 230  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 231  * initialized from the connection status.
 232  */
 233 static void ovs_ct_update_key(const struct sk_buff *skb,
 234                               const struct ovs_conntrack_info *info,
 235                               struct sw_flow_key *key, bool post_ct,
 236                               bool keep_nat_flags)
 237 {
 238         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 239         enum ip_conntrack_info ctinfo;
 240         struct nf_conn *ct;
 241         u8 state = 0;
 242
 243         ct = nf_ct_get(skb, &ctinfo);
 244         if (ct) {
 245                 state = ovs_ct_get_state(ctinfo);
 246                 /* All unconfirmed entries are NEW connections. */
 247                 if (!nf_ct_is_confirmed(ct))
 248                         state |= OVS_CS_F_NEW;
 249                 /* OVS persists the related flag for the duration of the
 250                  * connection.
 251                  */
 252                 if (ct->master)
 253                         state |= OVS_CS_F_RELATED;
 254                 if (keep_nat_flags) {
 255                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 256                 } else {
 257                         if (ct->status & IPS_SRC_NAT)
 258                                 state |= OVS_CS_F_SRC_NAT;
 259                         if (ct->status & IPS_DST_NAT)
 260                                 state |= OVS_CS_F_DST_NAT;
 261                 }
 262                 zone = nf_ct_zone(ct);
 263         } else if (post_ct) {
 264                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 265                 if (info)
 266                         zone = &info->zone;
 267         }
 268         __ovs_ct_update_key(key, state, zone, ct);
 269 }
 270
 271 /* This is called to initialize CT key fields possibly coming in from the local
 272  * stack.
 273  */
 274 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 275 {
 276         ovs_ct_update_key(skb, NULL, key, false, false);
 277 }
 278
 279 #define IN6_ADDR_INITIALIZER(ADDR) \
 280         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 281           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 282
 283 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 284                    const struct sw_flow_key *output, struct sk_buff *skb)
 285 {
 286         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 287                 return -EMSGSIZE;
 288
 289         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 290             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 291                 return -EMSGSIZE;
 292
 293         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 294             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 295                 return -EMSGSIZE;
 296
 297         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 298             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 299                     &output->ct.labels))
 300                 return -EMSGSIZE;
 301
 302         if (swkey->ct_orig_proto) {
 303                 if (swkey->eth.type == htons(ETH_P_IP)) {
 304                         struct ovs_key_ct_tuple_ipv4 orig = {
 305                                 output->ipv4.ct_orig.src,
 306                                 output->ipv4.ct_orig.dst,
 307                                 output->ct.orig_tp.src,
 308                                 output->ct.orig_tp.dst,
 309                                 output->ct_orig_proto,
 310                         };
 311                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 312                                     sizeof(orig), &orig))
 313                                 return -EMSGSIZE;
 314                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 315                         struct ovs_key_ct_tuple_ipv6 orig = {
 316                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 317                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 318                                 output->ct.orig_tp.src,
 319                                 output->ct.orig_tp.dst,
 320                                 output->ct_orig_proto,
 321                         };
 322                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 323                                     sizeof(orig), &orig))
 324                                 return -EMSGSIZE;
 325                 }
 326         }
 327
 328         return 0;
 329 }
 330
 331 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 332                            u32 ct_mark, u32 mask)
 333 {
 334 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 335         u32 new_mark;
 336
 337         new_mark = ct_mark | (ct->mark & ~(mask));
 338         if (ct->mark != new_mark) {
 339                 ct->mark = new_mark;
 340                 if (nf_ct_is_confirmed(ct))
 341                         nf_conntrack_event_cache(IPCT_MARK, ct);
 342                 key->ct.mark = new_mark;
 343         }
 344
 345         return 0;
 346 #else
 347         return -ENOTSUPP;
 348 #endif
 349 }
 350
 351 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 352 {
 353         struct nf_conn_labels *cl;
 354
 355         cl = nf_ct_labels_find(ct);
 356         if (!cl) {
 357                 nf_ct_labels_ext_add(ct);
 358                 cl = nf_ct_labels_find(ct);
 359         }
 360
 361         return cl;
 362 }
 363
 364 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 365  * since the new connection is not yet confirmed, and thus no-one else has
 366  * access to it's labels, we simply write them over.
 367  */
 368 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 369                               const struct ovs_key_ct_labels *labels,
 370                               const struct ovs_key_ct_labels *mask)
 371 {
 372         struct nf_conn_labels *cl, *master_cl;
 373         bool have_mask = labels_nonzero(mask);
 374
 375         /* Inherit master's labels to the related connection? */
 376         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 377
 378         if (!master_cl && !have_mask)
 379                 return 0;   /* Nothing to do. */
 380
 381         cl = ovs_ct_get_conn_labels(ct);
 382         if (!cl)
 383                 return -ENOSPC;
 384
 385         /* Inherit the master's labels, if any. */
 386         if (master_cl)
 387                 *cl = *master_cl;
 388
 389         if (have_mask) {
 390                 u32 *dst = (u32 *)cl->bits;
 391                 int i;
 392
 393                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 394                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 395                                 (labels->ct_labels_32[i]
 396                                  & mask->ct_labels_32[i]);
 397         }
 398
 399         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 400          * IPCT_LABEL bit is set in the event cache.
 401          */
 402         nf_conntrack_event_cache(IPCT_LABEL, ct);
 403
 404         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 405
 406         return 0;
 407 }
 408
 409 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 410                              const struct ovs_key_ct_labels *labels,
 411                              const struct ovs_key_ct_labels *mask)
 412 {
 413         struct nf_conn_labels *cl;
 414         int err;
 415
 416         cl = ovs_ct_get_conn_labels(ct);
 417         if (!cl)
 418                 return -ENOSPC;
 419
 420         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 421                                     mask->ct_labels_32,
 422                                     OVS_CT_LABELS_LEN_32);
 423         if (err)
 424                 return err;
 425
 426         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 427
 428         return 0;
 429 }
 430
 431 /* 'skb' should already be pulled to nh_ofs. */
 432 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 433 {
 434         const struct nf_conntrack_helper *helper;
 435         const struct nf_conn_help *help;
 436         enum ip_conntrack_info ctinfo;
 437         unsigned int protoff;
 438         struct nf_conn *ct;
 439         int err;
 440
 441         ct = nf_ct_get(skb, &ctinfo);
 442         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 443                 return NF_ACCEPT;
 444
 445         help = nfct_help(ct);
 446         if (!help)
 447                 return NF_ACCEPT;
 448
 449         helper = rcu_dereference(help->helper);
 450         if (!helper)
 451                 return NF_ACCEPT;
 452
 453         switch (proto) {
 454         case NFPROTO_IPV4:
 455                 protoff = ip_hdrlen(skb);
 456                 break;
 457         case NFPROTO_IPV6: {
 458                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 459                 __be16 frag_off;
 460                 int ofs;
 461
 462                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 463                                        &frag_off);
 464                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 465                         pr_debug("proto header not found\n");
 466                         return NF_ACCEPT;
 467                 }
 468                 protoff = ofs;
 469                 break;
 470         }
 471         default:
 472                 WARN_ONCE(1, "helper invoked on non-IP family!");
 473                 return NF_DROP;
 474         }
 475
 476         err = helper->help(skb, protoff, ct, ctinfo);
 477         if (err != NF_ACCEPT)
 478                 return err;
 479
 480         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 481          * FTP with NAT) adusting the TCP payload size when mangling IP
 482          * addresses and/or port numbers in the text-based control connection.
 483          */
 484         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 485             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 486                 return NF_DROP;
 487         return NF_ACCEPT;
 488 }
 489
 490 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 491  * value if 'skb' is freed.
 492  */
 493 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 494                             u16 zone, struct sk_buff *skb)
 495 {
 496         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 497         int err;
 498
 499         if (key->eth.type == htons(ETH_P_IP)) {
 500                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 501
 502                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 503                 err = ip_defrag(net, skb, user);
 504                 if (err)
 505                         return err;
 506
 507                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 508 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 509         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 510                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 511
 512                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 513                 err = nf_ct_frag6_gather(net, skb, user);
 514                 if (err) {
 515                         if (err != -EINPROGRESS)
 516                                 kfree_skb(skb);
 517                         return err;
 518                 }
 519
 520                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 521                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 522 #endif
 523         } else {
 524                 kfree_skb(skb);
 525                 return -EPFNOSUPPORT;
 526         }
 527
 528         /* The key extracted from the fragment that completed this datagram
 529          * likely didn't have an L4 header, so regenerate it.
 530          */
 531         ovs_flow_key_update_l3l4(skb, key);
 532
 533         key->ip.frag = OVS_FRAG_TYPE_NONE;
 534         skb_clear_hash(skb);
 535         skb->ignore_df = 1;
 536         *OVS_CB(skb) = ovs_cb;
 537
 538         return 0;
 539 }
 540
 541 static struct nf_conntrack_expect *
 542 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 543                    u16 proto, const struct sk_buff *skb)
 544 {
 545         struct nf_conntrack_tuple tuple;
 546         struct nf_conntrack_expect *exp;
 547
 548         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 549                 return NULL;
 550
 551         exp = __nf_ct_expect_find(net, zone, &tuple);
 552         if (exp) {
 553                 struct nf_conntrack_tuple_hash *h;
 554
 555                 /* Delete existing conntrack entry, if it clashes with the
 556                  * expectation.  This can happen since conntrack ALGs do not
 557                  * check for clashes between (new) expectations and existing
 558                  * conntrack entries.  nf_conntrack_in() will check the
 559                  * expectations only if a conntrack entry can not be found,
 560                  * which can lead to OVS finding the expectation (here) in the
 561                  * init direction, but which will not be removed by the
 562                  * nf_conntrack_in() call, if a matching conntrack entry is
 563                  * found instead.  In this case all init direction packets
 564                  * would be reported as new related packets, while reply
 565                  * direction packets would be reported as un-related
 566                  * established packets.
 567                  */
 568                 h = nf_conntrack_find_get(net, zone, &tuple);
 569                 if (h) {
 570                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 571
 572                         nf_ct_delete(ct, 0, 0);
 573                         nf_conntrack_put(&ct->ct_general);
 574                 }
 575         }
 576
 577         return exp;
 578 }
 579
 580 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 581 static enum ip_conntrack_info
 582 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 583 {
 584         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 585
 586         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 587                 return IP_CT_ESTABLISHED_REPLY;
 588         /* Once we've had two way comms, always ESTABLISHED. */
 589         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 590                 return IP_CT_ESTABLISHED;
 591         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 592                 return IP_CT_RELATED;
 593         return IP_CT_NEW;
 594 }
 595
 596 /* Find an existing connection which this packet belongs to without
 597  * re-attributing statistics or modifying the connection state.  This allows an
 598  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 599  *
 600  * Must be called with rcu_read_lock.
 601  *
 602  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 603  * if there is no existing entry.
 604  */
 605 static struct nf_conn *
 606 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 607                      u8 l3num, struct sk_buff *skb, bool natted)
 608 {
 609         struct nf_conntrack_tuple tuple;
 610         struct nf_conntrack_tuple_hash *h;
 611         struct nf_conn *ct;
 612
 613         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 614                                net, &tuple)) {
 615                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 616                 return NULL;
 617         }
 618
 619         /* Must invert the tuple if skb has been transformed by NAT. */
 620         if (natted) {
 621                 struct nf_conntrack_tuple inverse;
 622
 623                 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
 624                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 625                         return NULL;
 626                 }
 627                 tuple = inverse;
 628         }
 629
 630         /* look for tuple match */
 631         h = nf_conntrack_find_get(net, zone, &tuple);
 632         if (!h)
 633                 return NULL;   /* Not found. */
 634
 635         ct = nf_ct_tuplehash_to_ctrack(h);
 636
 637         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 638          * select the other tuplehash to get the right 'ctinfo' bits for this
 639          * packet.
 640          */
 641         if (natted)
 642                 h = &ct->tuplehash[!h->tuple.dst.dir];
 643
 644         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 645         return ct;
 646 }
 647
 648 static
 649 struct nf_conn *ovs_ct_executed(struct net *net,
 650                                 const struct sw_flow_key *key,
 651                                 const struct ovs_conntrack_info *info,
 652                                 struct sk_buff *skb,
 653                                 bool *ct_executed)
 654 {
 655         struct nf_conn *ct = NULL;
 656
 657         /* If no ct, check if we have evidence that an existing conntrack entry
 658          * might be found for this skb.  This happens when we lose a skb->_nfct
 659          * due to an upcall, or if the direction is being forced.  If the
 660          * connection was not confirmed, it is not cached and needs to be run
 661          * through conntrack again.
 662          */
 663         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 664                        !(key->ct_state & OVS_CS_F_INVALID) &&
 665                        (key->ct_zone == info->zone.id);
 666
 667         if (*ct_executed || (!key->ct_state && info->force)) {
 668                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 669                                           !!(key->ct_state &
 670                                           OVS_CS_F_NAT_MASK));
 671         }
 672
 673         return ct;
 674 }
 675
 676 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 677 static bool skb_nfct_cached(struct net *net,
 678                             const struct sw_flow_key *key,
 679                             const struct ovs_conntrack_info *info,
 680                             struct sk_buff *skb)
 681 {
 682         enum ip_conntrack_info ctinfo;
 683         struct nf_conn *ct;
 684         bool ct_executed = true;
 685
 686         ct = nf_ct_get(skb, &ctinfo);
 687         if (!ct)
 688                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 689
 690         if (ct)
 691                 nf_ct_get(skb, &ctinfo);
 692         else
 693                 return false;
 694
 695         if (!net_eq(net, read_pnet(&ct->ct_net)))
 696                 return false;
 697         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 698                 return false;
 699         if (info->helper) {
 700                 struct nf_conn_help *help;
 701
 702                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 703                 if (help && rcu_access_pointer(help->helper) != info->helper)
 704                         return false;
 705         }
 706         if (info->nf_ct_timeout) {
 707                 struct nf_conn_timeout *timeout_ext;
 708
 709                 timeout_ext = nf_ct_timeout_find(ct);
 710                 if (!timeout_ext || info->nf_ct_timeout !=
 711                     rcu_dereference(timeout_ext->timeout))
 712                         return false;
 713         }
 714         /* Force conntrack entry direction to the current packet? */
 715         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 716                 /* Delete the conntrack entry if confirmed, else just release
 717                  * the reference.
 718                  */
 719                 if (nf_ct_is_confirmed(ct))
 720                         nf_ct_delete(ct, 0, 0);
 721
 722                 nf_conntrack_put(&ct->ct_general);
 723                 nf_ct_set(skb, NULL, 0);
 724                 return false;
 725         }
 726
 727         return ct_executed;
 728 }
 729
 730 #if IS_ENABLED(CONFIG_NF_NAT)
 731 /* Modelled after nf_nat_ipv[46]_fn().
 732  * range is only used for new, uninitialized NAT state.
 733  * Returns either NF_ACCEPT or NF_DROP.
 734  */
 735 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 736                               enum ip_conntrack_info ctinfo,
 737                               const struct nf_nat_range2 *range,
 738                               enum nf_nat_manip_type maniptype)
 739 {
 740         int hooknum, nh_off, err = NF_ACCEPT;
 741
 742         nh_off = skb_network_offset(skb);
 743         skb_pull_rcsum(skb, nh_off);
 744
 745         /* See HOOK2MANIP(). */
 746         if (maniptype == NF_NAT_MANIP_SRC)
 747                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 748         else
 749                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 750
 751         switch (ctinfo) {
 752         case IP_CT_RELATED:
 753         case IP_CT_RELATED_REPLY:
 754                 if (IS_ENABLED(CONFIG_NF_NAT) &&
 755                     skb->protocol == htons(ETH_P_IP) &&
 756                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 757                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 758                                                            hooknum))
 759                                 err = NF_DROP;
 760                         goto push;
 761                 } else if (IS_ENABLED(CONFIG_IPV6) &&
 762                            skb->protocol == htons(ETH_P_IPV6)) {
 763                         __be16 frag_off;
 764                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 765                         int hdrlen = ipv6_skip_exthdr(skb,
 766                                                       sizeof(struct ipv6hdr),
 767                                                       &nexthdr, &frag_off);
 768
 769                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 770                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 771                                                                      ctinfo,
 772                                                                      hooknum,
 773                                                                      hdrlen))
 774                                         err = NF_DROP;
 775                                 goto push;
 776                         }
 777                 }
 778                 /* Non-ICMP, fall thru to initialize if needed. */
 779                 /* fall through */
 780         case IP_CT_NEW:
 781                 /* Seen it before?  This can happen for loopback, retrans,
 782                  * or local packets.
 783                  */
 784                 if (!nf_nat_initialized(ct, maniptype)) {
 785                         /* Initialize according to the NAT action. */
 786                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 787                                 /* Action is set up to establish a new
 788                                  * mapping.
 789                                  */
 790                                 ? nf_nat_setup_info(ct, range, maniptype)
 791                                 : nf_nat_alloc_null_binding(ct, hooknum);
 792                         if (err != NF_ACCEPT)
 793                                 goto push;
 794                 }
 795                 break;
 796
 797         case IP_CT_ESTABLISHED:
 798         case IP_CT_ESTABLISHED_REPLY:
 799                 break;
 800
 801         default:
 802                 err = NF_DROP;
 803                 goto push;
 804         }
 805
 806         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 807 push:
 808         skb_push(skb, nh_off);
 809         skb_postpush_rcsum(skb, skb->data, nh_off);
 810
 811         return err;
 812 }
 813
 814 static void ovs_nat_update_key(struct sw_flow_key *key,
 815                                const struct sk_buff *skb,
 816                                enum nf_nat_manip_type maniptype)
 817 {
 818         if (maniptype == NF_NAT_MANIP_SRC) {
 819                 __be16 src;
 820
 821                 key->ct_state |= OVS_CS_F_SRC_NAT;
 822                 if (key->eth.type == htons(ETH_P_IP))
 823                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 824                 else if (key->eth.type == htons(ETH_P_IPV6))
 825                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 826                                sizeof(key->ipv6.addr.src));
 827                 else
 828                         return;
 829
 830                 if (key->ip.proto == IPPROTO_UDP)
 831                         src = udp_hdr(skb)->source;
 832                 else if (key->ip.proto == IPPROTO_TCP)
 833                         src = tcp_hdr(skb)->source;
 834                 else if (key->ip.proto == IPPROTO_SCTP)
 835                         src = sctp_hdr(skb)->source;
 836                 else
 837                         return;
 838
 839                 key->tp.src = src;
 840         } else {
 841                 __be16 dst;
 842
 843                 key->ct_state |= OVS_CS_F_DST_NAT;
 844                 if (key->eth.type == htons(ETH_P_IP))
 845                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 846                 else if (key->eth.type == htons(ETH_P_IPV6))
 847                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 848                                sizeof(key->ipv6.addr.dst));
 849                 else
 850                         return;
 851
 852                 if (key->ip.proto == IPPROTO_UDP)
 853                         dst = udp_hdr(skb)->dest;
 854                 else if (key->ip.proto == IPPROTO_TCP)
 855                         dst = tcp_hdr(skb)->dest;
 856                 else if (key->ip.proto == IPPROTO_SCTP)
 857                         dst = sctp_hdr(skb)->dest;
 858                 else
 859                         return;
 860
 861                 key->tp.dst = dst;
 862         }
 863 }
 864
 865 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 866 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 867                       const struct ovs_conntrack_info *info,
 868                       struct sk_buff *skb, struct nf_conn *ct,
 869                       enum ip_conntrack_info ctinfo)
 870 {
 871         enum nf_nat_manip_type maniptype;
 872         int err;
 873
 874         /* Add NAT extension if not confirmed yet. */
 875         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 876                 return NF_ACCEPT;   /* Can't NAT. */
 877
 878         /* Determine NAT type.
 879          * Check if the NAT type can be deduced from the tracked connection.
 880          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 881          * when committing.
 882          */
 883         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 884             ct->status & IPS_NAT_MASK &&
 885             (ctinfo != IP_CT_RELATED || info->commit)) {
 886                 /* NAT an established or related connection like before. */
 887                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 888                         /* This is the REPLY direction for a connection
 889                          * for which NAT was applied in the forward
 890                          * direction.  Do the reverse NAT.
 891                          */
 892                         maniptype = ct->status & IPS_SRC_NAT
 893                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 894                 else
 895                         maniptype = ct->status & IPS_SRC_NAT
 896                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 897         } else if (info->nat & OVS_CT_SRC_NAT) {
 898                 maniptype = NF_NAT_MANIP_SRC;
 899         } else if (info->nat & OVS_CT_DST_NAT) {
 900                 maniptype = NF_NAT_MANIP_DST;
 901         } else {
 902                 return NF_ACCEPT; /* Connection is not NATed. */
 903         }
 904         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 905
 906         /* Mark NAT done if successful and update the flow key. */
 907         if (err == NF_ACCEPT)
 908                 ovs_nat_update_key(key, skb, maniptype);
 909
 910         return err;
 911 }
 912 #else /* !CONFIG_NF_NAT */
 913 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 914                       const struct ovs_conntrack_info *info,
 915                       struct sk_buff *skb, struct nf_conn *ct,
 916                       enum ip_conntrack_info ctinfo)
 917 {
 918         return NF_ACCEPT;
 919 }
 920 #endif
 921
 922 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 923  * not done already.  Update key with new CT state after passing the packet
 924  * through conntrack.
 925  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 926  * set to NULL and 0 will be returned.
 927  */
 928 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 929                            const struct ovs_conntrack_info *info,
 930                            struct sk_buff *skb)
 931 {
 932         /* If we are recirculating packets to match on conntrack fields and
 933          * committing with a separate conntrack action,  then we don't need to
 934          * actually run the packet through conntrack twice unless it's for a
 935          * different zone.
 936          */
 937         bool cached = skb_nfct_cached(net, key, info, skb);
 938         enum ip_conntrack_info ctinfo;
 939         struct nf_conn *ct;
 940
 941         if (!cached) {
 942                 struct nf_hook_state state = {
 943                         .hook = NF_INET_PRE_ROUTING,
 944                         .pf = info->family,
 945                         .net = net,
 946                 };
 947                 struct nf_conn *tmpl = info->ct;
 948                 int err;
 949
 950                 /* Associate skb with specified zone. */
 951                 if (tmpl) {
 952                         if (skb_nfct(skb))
 953                                 nf_conntrack_put(skb_nfct(skb));
 954                         nf_conntrack_get(&tmpl->ct_general);
 955                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 956                 }
 957
 958                 err = nf_conntrack_in(skb, &state);
 959                 if (err != NF_ACCEPT)
 960                         return -ENOENT;
 961
 962                 /* Clear CT state NAT flags to mark that we have not yet done
 963                  * NAT after the nf_conntrack_in() call.  We can actually clear
 964                  * the whole state, as it will be re-initialized below.
 965                  */
 966                 key->ct_state = 0;
 967
 968                 /* Update the key, but keep the NAT flags. */
 969                 ovs_ct_update_key(skb, info, key, true, true);
 970         }
 971
 972         ct = nf_ct_get(skb, &ctinfo);
 973         if (ct) {
 974                 /* Packets starting a new connection must be NATted before the
 975                  * helper, so that the helper knows about the NAT.  We enforce
 976                  * this by delaying both NAT and helper calls for unconfirmed
 977                  * connections until the committing CT action.  For later
 978                  * packets NAT and Helper may be called in either order.
 979                  *
 980                  * NAT will be done only if the CT action has NAT, and only
 981                  * once per packet (per zone), as guarded by the NAT bits in
 982                  * the key->ct_state.
 983                  */
 984                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 985                     (nf_ct_is_confirmed(ct) || info->commit) &&
 986                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 987                         return -EINVAL;
 988                 }
 989
 990                 /* Userspace may decide to perform a ct lookup without a helper
 991                  * specified followed by a (recirculate and) commit with one.
 992                  * Therefore, for unconfirmed connections which we will commit,
 993                  * we need to attach the helper here.
 994                  */
 995                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 996                     info->helper && !nfct_help(ct)) {
 997                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 998                                                             GFP_ATOMIC);
 999                         if (err)
1000                                 return err;
1001
1002                         /* helper installed, add seqadj if NAT is required */
1003                         if (info->nat && !nfct_seqadj(ct)) {
1004                                 if (!nfct_seqadj_ext_add(ct))
1005                                         return -EINVAL;
1006                         }
1007                 }
1008
1009                 /* Call the helper only if:
1010                  * - nf_conntrack_in() was executed above ("!cached") for a
1011                  *   confirmed connection, or
1012                  * - When committing an unconfirmed connection.
1013                  */
1014                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1015                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1016                         return -EINVAL;
1017                 }
1018         }
1019
1020         return 0;
1021 }
1022
1023 /* Lookup connection and read fields into key. */
1024 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1025                          const struct ovs_conntrack_info *info,
1026                          struct sk_buff *skb)
1027 {
1028         struct nf_conntrack_expect *exp;
1029
1030         /* If we pass an expected packet through nf_conntrack_in() the
1031          * expectation is typically removed, but the packet could still be
1032          * lost in upcall processing.  To prevent this from happening we
1033          * perform an explicit expectation lookup.  Expected connections are
1034          * always new, and will be passed through conntrack only when they are
1035          * committed, as it is OK to remove the expectation at that time.
1036          */
1037         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1038         if (exp) {
1039                 u8 state;
1040
1041                 /* NOTE: New connections are NATted and Helped only when
1042                  * committed, so we are not calling into NAT here.
1043                  */
1044                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1045                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1046         } else {
1047                 struct nf_conn *ct;
1048                 int err;
1049
1050                 err = __ovs_ct_lookup(net, key, info, skb);
1051                 if (err)
1052                         return err;
1053
1054                 ct = (struct nf_conn *)skb_nfct(skb);
1055                 if (ct)
1056                         nf_ct_deliver_cached_events(ct);
1057         }
1058
1059         return 0;
1060 }
1061
1062 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1063 {
1064         size_t i;
1065
1066         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1067                 if (labels->ct_labels_32[i])
1068                         return true;
1069
1070         return false;
1071 }
1072
1073 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1074 static struct hlist_head *ct_limit_hash_bucket(
1075         const struct ovs_ct_limit_info *info, u16 zone)
1076 {
1077         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1078 }
1079
1080 /* Call with ovs_mutex */
1081 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1082                          struct ovs_ct_limit *new_ct_limit)
1083 {
1084         struct ovs_ct_limit *ct_limit;
1085         struct hlist_head *head;
1086
1087         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1088         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1089                 if (ct_limit->zone == new_ct_limit->zone) {
1090                         hlist_replace_rcu(&ct_limit->hlist_node,
1091                                           &new_ct_limit->hlist_node);
1092                         kfree_rcu(ct_limit, rcu);
1093                         return;
1094                 }
1095         }
1096
1097         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1098 }
1099
1100 /* Call with ovs_mutex */
1101 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1102 {
1103         struct ovs_ct_limit *ct_limit;
1104         struct hlist_head *head;
1105         struct hlist_node *n;
1106
1107         head = ct_limit_hash_bucket(info, zone);
1108         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1109                 if (ct_limit->zone == zone) {
1110                         hlist_del_rcu(&ct_limit->hlist_node);
1111                         kfree_rcu(ct_limit, rcu);
1112                         return;
1113                 }
1114         }
1115 }
1116
1117 /* Call with RCU read lock */
1118 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1119 {
1120         struct ovs_ct_limit *ct_limit;
1121         struct hlist_head *head;
1122
1123         head = ct_limit_hash_bucket(info, zone);
1124         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1125                 if (ct_limit->zone == zone)
1126                         return ct_limit->limit;
1127         }
1128
1129         return info->default_limit;
1130 }
1131
1132 static int ovs_ct_check_limit(struct net *net,
1133                               const struct ovs_conntrack_info *info,
1134                               const struct nf_conntrack_tuple *tuple)
1135 {
1136         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1137         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1138         u32 per_zone_limit, connections;
1139         u32 conncount_key;
1140
1141         conncount_key = info->zone.id;
1142
1143         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1144         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1145                 return 0;
1146
1147         connections = nf_conncount_count(net, ct_limit_info->data,
1148                                          &conncount_key, tuple, &info->zone);
1149         if (connections > per_zone_limit)
1150                 return -ENOMEM;
1151
1152         return 0;
1153 }
1154 #endif
1155
1156 /* Lookup connection and confirm if unconfirmed. */
1157 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1158                          const struct ovs_conntrack_info *info,
1159                          struct sk_buff *skb)
1160 {
1161         enum ip_conntrack_info ctinfo;
1162         struct nf_conn *ct;
1163         int err;
1164
1165         err = __ovs_ct_lookup(net, key, info, skb);
1166         if (err)
1167                 return err;
1168
1169         /* The connection could be invalid, in which case this is a no-op.*/
1170         ct = nf_ct_get(skb, &ctinfo);
1171         if (!ct)
1172                 return 0;
1173
1174 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1175         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1176                 if (!nf_ct_is_confirmed(ct)) {
1177                         err = ovs_ct_check_limit(net, info,
1178                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1179                         if (err) {
1180                                 net_warn_ratelimited("openvswitch: zone: %u "
1181                                         "exceeds conntrack limit\n",
1182                                         info->zone.id);
1183                                 return err;
1184                         }
1185                 }
1186         }
1187 #endif
1188
1189         /* Set the conntrack event mask if given.  NEW and DELETE events have
1190          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1191          * typically would receive many kinds of updates.  Setting the event
1192          * mask allows those events to be filtered.  The set event mask will
1193          * remain in effect for the lifetime of the connection unless changed
1194          * by a further CT action with both the commit flag and the eventmask
1195          * option. */
1196         if (info->have_eventmask) {
1197                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1198
1199                 if (cache)
1200                         cache->ctmask = info->eventmask;
1201         }
1202
1203         /* Apply changes before confirming the connection so that the initial
1204          * conntrack NEW netlink event carries the values given in the CT
1205          * action.
1206          */
1207         if (info->mark.mask) {
1208                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1209                                       info->mark.mask);
1210                 if (err)
1211                         return err;
1212         }
1213         if (!nf_ct_is_confirmed(ct)) {
1214                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1215                                          &info->labels.mask);
1216                 if (err)
1217                         return err;
1218         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1219                    labels_nonzero(&info->labels.mask)) {
1220                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1221                                         &info->labels.mask);
1222                 if (err)
1223                         return err;
1224         }
1225         /* This will take care of sending queued events even if the connection
1226          * is already confirmed.
1227          */
1228         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1229                 return -EINVAL;
1230
1231         return 0;
1232 }
1233
1234 /* Trim the skb to the length specified by the IP/IPv6 header,
1235  * removing any trailing lower-layer padding. This prepares the skb
1236  * for higher-layer processing that assumes skb->len excludes padding
1237  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1238  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1239  */
1240 static int ovs_skb_network_trim(struct sk_buff *skb)
1241 {
1242         unsigned int len;
1243         int err;
1244
1245         switch (skb->protocol) {
1246         case htons(ETH_P_IP):
1247                 len = ntohs(ip_hdr(skb)->tot_len);
1248                 break;
1249         case htons(ETH_P_IPV6):
1250                 len = sizeof(struct ipv6hdr)
1251                         + ntohs(ipv6_hdr(skb)->payload_len);
1252                 break;
1253         default:
1254                 len = skb->len;
1255         }
1256
1257         err = pskb_trim_rcsum(skb, len);
1258         if (err)
1259                 kfree_skb(skb);
1260
1261         return err;
1262 }
1263
1264 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1265  * value if 'skb' is freed.
1266  */
1267 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1268                    struct sw_flow_key *key,
1269                    const struct ovs_conntrack_info *info)
1270 {
1271         int nh_ofs;
1272         int err;
1273
1274         /* The conntrack module expects to be working at L3. */
1275         nh_ofs = skb_network_offset(skb);
1276         skb_pull_rcsum(skb, nh_ofs);
1277
1278         err = ovs_skb_network_trim(skb);
1279         if (err)
1280                 return err;
1281
1282         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1283                 err = handle_fragments(net, key, info->zone.id, skb);
1284                 if (err)
1285                         return err;
1286         }
1287
1288         if (info->commit)
1289                 err = ovs_ct_commit(net, key, info, skb);
1290         else
1291                 err = ovs_ct_lookup(net, key, info, skb);
1292
1293         skb_push(skb, nh_ofs);
1294         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1295         if (err)
1296                 kfree_skb(skb);
1297         return err;
1298 }
1299
1300 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1301 {
1302         if (skb_nfct(skb)) {
1303                 nf_conntrack_put(skb_nfct(skb));
1304                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1305                 ovs_ct_fill_key(skb, key);
1306         }
1307
1308         return 0;
1309 }
1310
1311 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1312                              const struct sw_flow_key *key, bool log)
1313 {
1314         struct nf_conntrack_helper *helper;
1315         struct nf_conn_help *help;
1316         int ret = 0;
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 #if IS_ENABLED(CONFIG_NF_NAT)
1332         if (info->nat) {
1333                 ret = nf_nat_helper_try_module_get(name, info->family,
1334                                                    key->ip.proto);
1335                 if (ret) {
1336                         nf_conntrack_helper_put(helper);
1337                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1338                                   name, ret);
1339                         return ret;
1340                 }
1341         }
1342 #endif
1343         rcu_assign_pointer(help->helper, helper);
1344         info->helper = helper;
1345         return ret;
1346 }
1347
1348 #if IS_ENABLED(CONFIG_NF_NAT)
1349 static int parse_nat(const struct nlattr *attr,
1350                      struct ovs_conntrack_info *info, bool log)
1351 {
1352         struct nlattr *a;
1353         int rem;
1354         bool have_ip_max = false;
1355         bool have_proto_max = false;
1356         bool ip_vers = (info->family == NFPROTO_IPV6);
1357
1358         nla_for_each_nested(a, attr, rem) {
1359                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1360                         [OVS_NAT_ATTR_SRC] = {0, 0},
1361                         [OVS_NAT_ATTR_DST] = {0, 0},
1362                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1363                                                  sizeof(struct in6_addr)},
1364                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1365                                                  sizeof(struct in6_addr)},
1366                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1367                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1368                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1369                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1370                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1371                 };
1372                 int type = nla_type(a);
1373
1374                 if (type > OVS_NAT_ATTR_MAX) {
1375                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1376                                   type, OVS_NAT_ATTR_MAX);
1377                         return -EINVAL;
1378                 }
1379
1380                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1381                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1382                                   type, nla_len(a),
1383                                   ovs_nat_attr_lens[type][ip_vers]);
1384                         return -EINVAL;
1385                 }
1386
1387                 switch (type) {
1388                 case OVS_NAT_ATTR_SRC:
1389                 case OVS_NAT_ATTR_DST:
1390                         if (info->nat) {
1391                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1392                                 return -ERANGE;
1393                         }
1394                         info->nat |= OVS_CT_NAT;
1395                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1396                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1397                         break;
1398
1399                 case OVS_NAT_ATTR_IP_MIN:
1400                         nla_memcpy(&info->range.min_addr, a,
1401                                    sizeof(info->range.min_addr));
1402                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1403                         break;
1404
1405                 case OVS_NAT_ATTR_IP_MAX:
1406                         have_ip_max = true;
1407                         nla_memcpy(&info->range.max_addr, a,
1408                                    sizeof(info->range.max_addr));
1409                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1410                         break;
1411
1412                 case OVS_NAT_ATTR_PROTO_MIN:
1413                         info->range.min_proto.all = htons(nla_get_u16(a));
1414                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1415                         break;
1416
1417                 case OVS_NAT_ATTR_PROTO_MAX:
1418                         have_proto_max = true;
1419                         info->range.max_proto.all = htons(nla_get_u16(a));
1420                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1421                         break;
1422
1423                 case OVS_NAT_ATTR_PERSISTENT:
1424                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1425                         break;
1426
1427                 case OVS_NAT_ATTR_PROTO_HASH:
1428                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1429                         break;
1430
1431                 case OVS_NAT_ATTR_PROTO_RANDOM:
1432                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1433                         break;
1434
1435                 default:
1436                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1437                         return -EINVAL;
1438                 }
1439         }
1440
1441         if (rem > 0) {
1442                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1443                 return -EINVAL;
1444         }
1445         if (!info->nat) {
1446                 /* Do not allow flags if no type is given. */
1447                 if (info->range.flags) {
1448                         OVS_NLERR(log,
1449                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1450                                   );
1451                         return -EINVAL;
1452                 }
1453                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1454         } else if (!info->commit) {
1455                 OVS_NLERR(log,
1456                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1457                           );
1458                 return -EINVAL;
1459         }
1460         /* Allow missing IP_MAX. */
1461         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1462                 memcpy(&info->range.max_addr, &info->range.min_addr,
1463                        sizeof(info->range.max_addr));
1464         }
1465         /* Allow missing PROTO_MAX. */
1466         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1467             !have_proto_max) {
1468                 info->range.max_proto.all = info->range.min_proto.all;
1469         }
1470         return 0;
1471 }
1472 #endif
1473
1474 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1475         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1476         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1477         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1478                                     .maxlen = sizeof(u16) },
1479         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1480                                     .maxlen = sizeof(struct md_mark) },
1481         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1482                                     .maxlen = sizeof(struct md_labels) },
1483         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1484                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1485 #if IS_ENABLED(CONFIG_NF_NAT)
1486         /* NAT length is checked when parsing the nested attributes. */
1487         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1488 #endif
1489         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1490                                     .maxlen = sizeof(u32) },
1491         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1492                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1493 };
1494
1495 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1496                     const char **helper, bool log)
1497 {
1498         struct nlattr *a;
1499         int rem;
1500
1501         nla_for_each_nested(a, attr, rem) {
1502                 int type = nla_type(a);
1503                 int maxlen;
1504                 int minlen;
1505
1506                 if (type > OVS_CT_ATTR_MAX) {
1507                         OVS_NLERR(log,
1508                                   "Unknown conntrack attr (type=%d, max=%d)",
1509                                   type, OVS_CT_ATTR_MAX);
1510                         return -EINVAL;
1511                 }
1512
1513                 maxlen = ovs_ct_attr_lens[type].maxlen;
1514                 minlen = ovs_ct_attr_lens[type].minlen;
1515                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1516                         OVS_NLERR(log,
1517                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1518                                   type, nla_len(a), maxlen);
1519                         return -EINVAL;
1520                 }
1521
1522                 switch (type) {
1523                 case OVS_CT_ATTR_FORCE_COMMIT:
1524                         info->force = true;
1525                         /* fall through. */
1526                 case OVS_CT_ATTR_COMMIT:
1527                         info->commit = true;
1528                         break;
1529 #ifdef CONFIG_NF_CONNTRACK_ZONES
1530                 case OVS_CT_ATTR_ZONE:
1531                         info->zone.id = nla_get_u16(a);
1532                         break;
1533 #endif
1534 #ifdef CONFIG_NF_CONNTRACK_MARK
1535                 case OVS_CT_ATTR_MARK: {
1536                         struct md_mark *mark = nla_data(a);
1537
1538                         if (!mark->mask) {
1539                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1540                                 return -EINVAL;
1541                         }
1542                         info->mark = *mark;
1543                         break;
1544                 }
1545 #endif
1546 #ifdef CONFIG_NF_CONNTRACK_LABELS
1547                 case OVS_CT_ATTR_LABELS: {
1548                         struct md_labels *labels = nla_data(a);
1549
1550                         if (!labels_nonzero(&labels->mask)) {
1551                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1552                                 return -EINVAL;
1553                         }
1554                         info->labels = *labels;
1555                         break;
1556                 }
1557 #endif
1558                 case OVS_CT_ATTR_HELPER:
1559                         *helper = nla_data(a);
1560                         if (!memchr(*helper, '\0', nla_len(a))) {
1561                                 OVS_NLERR(log, "Invalid conntrack helper");
1562                                 return -EINVAL;
1563                         }
1564                         break;
1565 #if IS_ENABLED(CONFIG_NF_NAT)
1566                 case OVS_CT_ATTR_NAT: {
1567                         int err = parse_nat(a, info, log);
1568
1569                         if (err)
1570                                 return err;
1571                         break;
1572                 }
1573 #endif
1574                 case OVS_CT_ATTR_EVENTMASK:
1575                         info->have_eventmask = true;
1576                         info->eventmask = nla_get_u32(a);
1577                         break;
1578 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1579                 case OVS_CT_ATTR_TIMEOUT:
1580                         memcpy(info->timeout, nla_data(a), nla_len(a));
1581                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1582                                 OVS_NLERR(log, "Invalid conntrack timeout");
1583                                 return -EINVAL;
1584                         }
1585                         break;
1586 #endif
1587
1588                 default:
1589                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1590                                   type);
1591                         return -EINVAL;
1592                 }
1593         }
1594
1595 #ifdef CONFIG_NF_CONNTRACK_MARK
1596         if (!info->commit && info->mark.mask) {
1597                 OVS_NLERR(log,
1598                           "Setting conntrack mark requires 'commit' flag.");
1599                 return -EINVAL;
1600         }
1601 #endif
1602 #ifdef CONFIG_NF_CONNTRACK_LABELS
1603         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1604                 OVS_NLERR(log,
1605                           "Setting conntrack labels requires 'commit' flag.");
1606                 return -EINVAL;
1607         }
1608 #endif
1609         if (rem > 0) {
1610                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1611                 return -EINVAL;
1612         }
1613
1614         return 0;
1615 }
1616
1617 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1618 {
1619         if (attr == OVS_KEY_ATTR_CT_STATE)
1620                 return true;
1621         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1622             attr == OVS_KEY_ATTR_CT_ZONE)
1623                 return true;
1624         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1625             attr == OVS_KEY_ATTR_CT_MARK)
1626                 return true;
1627         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1628             attr == OVS_KEY_ATTR_CT_LABELS) {
1629                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1630
1631                 return ovs_net->xt_label;
1632         }
1633
1634         return false;
1635 }
1636
1637 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1638                        const struct sw_flow_key *key,
1639                        struct sw_flow_actions **sfa,  bool log)
1640 {
1641         struct ovs_conntrack_info ct_info;
1642         const char *helper = NULL;
1643         u16 family;
1644         int err;
1645
1646         family = key_to_nfproto(key);
1647         if (family == NFPROTO_UNSPEC) {
1648                 OVS_NLERR(log, "ct family unspecified");
1649                 return -EINVAL;
1650         }
1651
1652         memset(&ct_info, 0, sizeof(ct_info));
1653         ct_info.family = family;
1654
1655         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1656                         NF_CT_DEFAULT_ZONE_DIR, 0);
1657
1658         err = parse_ct(attr, &ct_info, &helper, log);
1659         if (err)
1660                 return err;
1661
1662         /* Set up template for tracking connections in specific zones. */
1663         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1664         if (!ct_info.ct) {
1665                 OVS_NLERR(log, "Failed to allocate conntrack template");
1666                 return -ENOMEM;
1667         }
1668
1669         if (ct_info.timeout[0]) {
1670                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1671                                       ct_info.timeout))
1672                         pr_info_ratelimited("Failed to associated timeout "
1673                                             "policy `%s'\n", ct_info.timeout);
1674                 else
1675                         ct_info.nf_ct_timeout = rcu_dereference(
1676                                 nf_ct_timeout_find(ct_info.ct)->timeout);
1677
1678         }
1679
1680         if (helper) {
1681                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1682                 if (err)
1683                         goto err_free_ct;
1684         }
1685
1686         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1687                                  sizeof(ct_info), log);
1688         if (err)
1689                 goto err_free_ct;
1690
1691         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1692         nf_conntrack_get(&ct_info.ct->ct_general);
1693         return 0;
1694 err_free_ct:
1695         __ovs_ct_free_action(&ct_info);
1696         return err;
1697 }
1698
1699 #if IS_ENABLED(CONFIG_NF_NAT)
1700 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1701                                struct sk_buff *skb)
1702 {
1703         struct nlattr *start;
1704
1705         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1706         if (!start)
1707                 return false;
1708
1709         if (info->nat & OVS_CT_SRC_NAT) {
1710                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1711                         return false;
1712         } else if (info->nat & OVS_CT_DST_NAT) {
1713                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1714                         return false;
1715         } else {
1716                 goto out;
1717         }
1718
1719         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1720                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1721                     info->family == NFPROTO_IPV4) {
1722                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1723                                             info->range.min_addr.ip) ||
1724                             (info->range.max_addr.ip
1725                              != info->range.min_addr.ip &&
1726                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1727                                               info->range.max_addr.ip))))
1728                                 return false;
1729                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1730                            info->family == NFPROTO_IPV6) {
1731                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1732                                              &info->range.min_addr.in6) ||
1733                             (memcmp(&info->range.max_addr.in6,
1734                                     &info->range.min_addr.in6,
1735                                     sizeof(info->range.max_addr.in6)) &&
1736                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1737                                                &info->range.max_addr.in6))))
1738                                 return false;
1739                 } else {
1740                         return false;
1741                 }
1742         }
1743         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1744             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1745                          ntohs(info->range.min_proto.all)) ||
1746              (info->range.max_proto.all != info->range.min_proto.all &&
1747               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1748                           ntohs(info->range.max_proto.all)))))
1749                 return false;
1750
1751         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1752             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1753                 return false;
1754         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1755             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1756                 return false;
1757         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1758             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1759                 return false;
1760 out:
1761         nla_nest_end(skb, start);
1762
1763         return true;
1764 }
1765 #endif
1766
1767 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1768                           struct sk_buff *skb)
1769 {
1770         struct nlattr *start;
1771
1772         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1773         if (!start)
1774                 return -EMSGSIZE;
1775
1776         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1777                                             ? OVS_CT_ATTR_FORCE_COMMIT
1778                                             : OVS_CT_ATTR_COMMIT))
1779                 return -EMSGSIZE;
1780         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1781             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1782                 return -EMSGSIZE;
1783         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1784             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1785                     &ct_info->mark))
1786                 return -EMSGSIZE;
1787         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1788             labels_nonzero(&ct_info->labels.mask) &&
1789             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1790                     &ct_info->labels))
1791                 return -EMSGSIZE;
1792         if (ct_info->helper) {
1793                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1794                                    ct_info->helper->name))
1795                         return -EMSGSIZE;
1796         }
1797         if (ct_info->have_eventmask &&
1798             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1799                 return -EMSGSIZE;
1800         if (ct_info->timeout[0]) {
1801                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1802                         return -EMSGSIZE;
1803         }
1804
1805 #if IS_ENABLED(CONFIG_NF_NAT)
1806         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1807                 return -EMSGSIZE;
1808 #endif
1809         nla_nest_end(skb, start);
1810
1811         return 0;
1812 }
1813
1814 void ovs_ct_free_action(const struct nlattr *a)
1815 {
1816         struct ovs_conntrack_info *ct_info = nla_data(a);
1817
1818         __ovs_ct_free_action(ct_info);
1819 }
1820
1821 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1822 {
1823         if (ct_info->helper) {
1824 #if IS_ENABLED(CONFIG_NF_NAT)
1825                 if (ct_info->nat)
1826                         nf_nat_helper_put(ct_info->helper);
1827 #endif
1828                 nf_conntrack_helper_put(ct_info->helper);
1829         }
1830         if (ct_info->ct) {
1831                 if (ct_info->timeout[0])
1832                         nf_ct_destroy_timeout(ct_info->ct);
1833                 nf_ct_tmpl_free(ct_info->ct);
1834         }
1835 }
1836
1837 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1838 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1839 {
1840         int i, err;
1841
1842         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1843                                          GFP_KERNEL);
1844         if (!ovs_net->ct_limit_info)
1845                 return -ENOMEM;
1846
1847         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1848         ovs_net->ct_limit_info->limits =
1849                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1850                               GFP_KERNEL);
1851         if (!ovs_net->ct_limit_info->limits) {
1852                 kfree(ovs_net->ct_limit_info);
1853                 return -ENOMEM;
1854         }
1855
1856         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1857                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1858
1859         ovs_net->ct_limit_info->data =
1860                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1861
1862         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1863                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1864                 kfree(ovs_net->ct_limit_info->limits);
1865                 kfree(ovs_net->ct_limit_info);
1866                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1867                 return err;
1868         }
1869         return 0;
1870 }
1871
1872 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1873 {
1874         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1875         int i;
1876
1877         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1878         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1879                 struct hlist_head *head = &info->limits[i];
1880                 struct ovs_ct_limit *ct_limit;
1881
1882                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1883                         kfree_rcu(ct_limit, rcu);
1884         }
1885         kfree(ovs_net->ct_limit_info->limits);
1886         kfree(ovs_net->ct_limit_info);
1887 }
1888
1889 static struct sk_buff *
1890 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1891                              struct ovs_header **ovs_reply_header)
1892 {
1893         struct ovs_header *ovs_header = info->userhdr;
1894         struct sk_buff *skb;
1895
1896         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1897         if (!skb)
1898                 return ERR_PTR(-ENOMEM);
1899
1900         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1901                                         info->snd_seq,
1902                                         &dp_ct_limit_genl_family, 0, cmd);
1903
1904         if (!*ovs_reply_header) {
1905                 nlmsg_free(skb);
1906                 return ERR_PTR(-EMSGSIZE);
1907         }
1908         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1909
1910         return skb;
1911 }
1912
1913 static bool check_zone_id(int zone_id, u16 *pzone)
1914 {
1915         if (zone_id >= 0 && zone_id <= 65535) {
1916                 *pzone = (u16)zone_id;
1917                 return true;
1918         }
1919         return false;
1920 }
1921
1922 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1923                                        struct ovs_ct_limit_info *info)
1924 {
1925         struct ovs_zone_limit *zone_limit;
1926         int rem;
1927         u16 zone;
1928
1929         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1930         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1931
1932         while (rem >= sizeof(*zone_limit)) {
1933                 if (unlikely(zone_limit->zone_id ==
1934                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1935                         ovs_lock();
1936                         info->default_limit = zone_limit->limit;
1937                         ovs_unlock();
1938                 } else if (unlikely(!check_zone_id(
1939                                 zone_limit->zone_id, &zone))) {
1940                         OVS_NLERR(true, "zone id is out of range");
1941                 } else {
1942                         struct ovs_ct_limit *ct_limit;
1943
1944                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1945                         if (!ct_limit)
1946                                 return -ENOMEM;
1947
1948                         ct_limit->zone = zone;
1949                         ct_limit->limit = zone_limit->limit;
1950
1951                         ovs_lock();
1952                         ct_limit_set(info, ct_limit);
1953                         ovs_unlock();
1954                 }
1955                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1956                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1957                                 NLA_ALIGN(sizeof(*zone_limit)));
1958         }
1959
1960         if (rem)
1961                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1962
1963         return 0;
1964 }
1965
1966 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1967                                        struct ovs_ct_limit_info *info)
1968 {
1969         struct ovs_zone_limit *zone_limit;
1970         int rem;
1971         u16 zone;
1972
1973         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1974         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1975
1976         while (rem >= sizeof(*zone_limit)) {
1977                 if (unlikely(zone_limit->zone_id ==
1978                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1979                         ovs_lock();
1980                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1981                         ovs_unlock();
1982                 } else if (unlikely(!check_zone_id(
1983                                 zone_limit->zone_id, &zone))) {
1984                         OVS_NLERR(true, "zone id is out of range");
1985                 } else {
1986                         ovs_lock();
1987                         ct_limit_del(info, zone);
1988                         ovs_unlock();
1989                 }
1990                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1991                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1992                                 NLA_ALIGN(sizeof(*zone_limit)));
1993         }
1994
1995         if (rem)
1996                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1997
1998         return 0;
1999 }
2000
2001 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2002                                           struct sk_buff *reply)
2003 {
2004         struct ovs_zone_limit zone_limit;
2005         int err;
2006
2007         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2008         zone_limit.limit = info->default_limit;
2009         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2010         if (err)
2011                 return err;
2012
2013         return 0;
2014 }
2015
2016 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2017                                          struct nf_conncount_data *data,
2018                                          u16 zone_id, u32 limit,
2019                                          struct sk_buff *reply)
2020 {
2021         struct nf_conntrack_zone ct_zone;
2022         struct ovs_zone_limit zone_limit;
2023         u32 conncount_key = zone_id;
2024
2025         zone_limit.zone_id = zone_id;
2026         zone_limit.limit = limit;
2027         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2028
2029         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2030                                               &ct_zone);
2031         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2032 }
2033
2034 static int ovs_ct_limit_get_zone_limit(struct net *net,
2035                                        struct nlattr *nla_zone_limit,
2036                                        struct ovs_ct_limit_info *info,
2037                                        struct sk_buff *reply)
2038 {
2039         struct ovs_zone_limit *zone_limit;
2040         int rem, err;
2041         u32 limit;
2042         u16 zone;
2043
2044         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2045         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2046
2047         while (rem >= sizeof(*zone_limit)) {
2048                 if (unlikely(zone_limit->zone_id ==
2049                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2050                         err = ovs_ct_limit_get_default_limit(info, reply);
2051                         if (err)
2052                                 return err;
2053                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2054                                                         &zone))) {
2055                         OVS_NLERR(true, "zone id is out of range");
2056                 } else {
2057                         rcu_read_lock();
2058                         limit = ct_limit_get(info, zone);
2059                         rcu_read_unlock();
2060
2061                         err = __ovs_ct_limit_get_zone_limit(
2062                                 net, info->data, zone, limit, reply);
2063                         if (err)
2064                                 return err;
2065                 }
2066                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2067                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2068                                 NLA_ALIGN(sizeof(*zone_limit)));
2069         }
2070
2071         if (rem)
2072                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2073
2074         return 0;
2075 }
2076
2077 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2078                                            struct ovs_ct_limit_info *info,
2079                                            struct sk_buff *reply)
2080 {
2081         struct ovs_ct_limit *ct_limit;
2082         struct hlist_head *head;
2083         int i, err = 0;
2084
2085         err = ovs_ct_limit_get_default_limit(info, reply);
2086         if (err)
2087                 return err;
2088
2089         rcu_read_lock();
2090         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2091                 head = &info->limits[i];
2092                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2093                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2094                                 ct_limit->zone, ct_limit->limit, reply);
2095                         if (err)
2096                                 goto exit_err;
2097                 }
2098         }
2099
2100 exit_err:
2101         rcu_read_unlock();
2102         return err;
2103 }
2104
2105 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2106 {
2107         struct nlattr **a = info->attrs;
2108         struct sk_buff *reply;
2109         struct ovs_header *ovs_reply_header;
2110         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2111         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2112         int err;
2113
2114         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2115                                              &ovs_reply_header);
2116         if (IS_ERR(reply))
2117                 return PTR_ERR(reply);
2118
2119         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2120                 err = -EINVAL;
2121                 goto exit_err;
2122         }
2123
2124         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2125                                           ct_limit_info);
2126         if (err)
2127                 goto exit_err;
2128
2129         static_branch_enable(&ovs_ct_limit_enabled);
2130
2131         genlmsg_end(reply, ovs_reply_header);
2132         return genlmsg_reply(reply, info);
2133
2134 exit_err:
2135         nlmsg_free(reply);
2136         return err;
2137 }
2138
2139 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2140 {
2141         struct nlattr **a = info->attrs;
2142         struct sk_buff *reply;
2143         struct ovs_header *ovs_reply_header;
2144         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2145         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2146         int err;
2147
2148         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2149                                              &ovs_reply_header);
2150         if (IS_ERR(reply))
2151                 return PTR_ERR(reply);
2152
2153         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2154                 err = -EINVAL;
2155                 goto exit_err;
2156         }
2157
2158         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2159                                           ct_limit_info);
2160         if (err)
2161                 goto exit_err;
2162
2163         genlmsg_end(reply, ovs_reply_header);
2164         return genlmsg_reply(reply, info);
2165
2166 exit_err:
2167         nlmsg_free(reply);
2168         return err;
2169 }
2170
2171 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2172 {
2173         struct nlattr **a = info->attrs;
2174         struct nlattr *nla_reply;
2175         struct sk_buff *reply;
2176         struct ovs_header *ovs_reply_header;
2177         struct net *net = sock_net(skb->sk);
2178         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2179         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2180         int err;
2181
2182         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2183                                              &ovs_reply_header);
2184         if (IS_ERR(reply))
2185                 return PTR_ERR(reply);
2186
2187         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2188         if (!nla_reply) {
2189                 err = -EMSGSIZE;
2190                 goto exit_err;
2191         }
2192
2193         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2194                 err = ovs_ct_limit_get_zone_limit(
2195                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2196                         reply);
2197                 if (err)
2198                         goto exit_err;
2199         } else {
2200                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2201                                                       reply);
2202                 if (err)
2203                         goto exit_err;
2204         }
2205
2206         nla_nest_end(reply, nla_reply);
2207         genlmsg_end(reply, ovs_reply_header);
2208         return genlmsg_reply(reply, info);
2209
2210 exit_err:
2211         nlmsg_free(reply);
2212         return err;
2213 }
2214
2215 static struct genl_ops ct_limit_genl_ops[] = {
2216         { .cmd = OVS_CT_LIMIT_CMD_SET,
2217                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2218                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2219                                            * privilege. */
2220                 .doit = ovs_ct_limit_cmd_set,
2221         },
2222         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2223                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2224                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2225                                            * privilege. */
2226                 .doit = ovs_ct_limit_cmd_del,
2227         },
2228         { .cmd = OVS_CT_LIMIT_CMD_GET,
2229                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2230                 .flags = 0,               /* OK for unprivileged users. */
2231                 .doit = ovs_ct_limit_cmd_get,
2232         },
2233 };
2234
2235 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2236         .name = OVS_CT_LIMIT_MCGROUP,
2237 };
2238
2239 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2240         .hdrsize = sizeof(struct ovs_header),
2241         .name = OVS_CT_LIMIT_FAMILY,
2242         .version = OVS_CT_LIMIT_VERSION,
2243         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2244         .policy = ct_limit_policy,
2245         .netnsok = true,
2246         .parallel_ops = true,
2247         .ops = ct_limit_genl_ops,
2248         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2249         .mcgrps = &ovs_ct_limit_multicast_group,
2250         .n_mcgrps = 1,
2251         .module = THIS_MODULE,
2252 };
2253 #endif
2254
2255 int ovs_ct_init(struct net *net)
2256 {
2257         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2258         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2259
2260         if (nf_connlabels_get(net, n_bits - 1)) {
2261                 ovs_net->xt_label = false;
2262                 OVS_NLERR(true, "Failed to set connlabel length");
2263         } else {
2264                 ovs_net->xt_label = true;
2265         }
2266
2267 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2268         return ovs_ct_limit_init(net, ovs_net);
2269 #else
2270         return 0;
2271 #endif
2272 }
2273
2274 void ovs_ct_exit(struct net *net)
2275 {
2276         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2277
2278 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2279         ovs_ct_limit_exit(net, ovs_net);
2280 #endif
2281
2282         if (ovs_net->xt_label)
2283                 nf_connlabels_put(net);
2284 }