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                 bool add_helper = false;
 975
 976                 /* Packets starting a new connection must be NATted before the
 977                  * helper, so that the helper knows about the NAT.  We enforce
 978                  * this by delaying both NAT and helper calls for unconfirmed
 979                  * connections until the committing CT action.  For later
 980                  * packets NAT and Helper may be called in either order.
 981                  *
 982                  * NAT will be done only if the CT action has NAT, and only
 983                  * once per packet (per zone), as guarded by the NAT bits in
 984                  * the key->ct_state.
 985                  */
 986                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 987                     (nf_ct_is_confirmed(ct) || info->commit) &&
 988                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 989                         return -EINVAL;
 990                 }
 991
 992                 /* Userspace may decide to perform a ct lookup without a helper
 993                  * specified followed by a (recirculate and) commit with one,
 994                  * or attach a helper in a later commit.  Therefore, for
 995                  * connections which we will commit, we may need to attach
 996                  * the helper here.
 997                  */
 998                 if (info->commit && info->helper && !nfct_help(ct)) {
 999                         int err = __nf_ct_try_assign_helper(ct, info->ct,
1000                                                             GFP_ATOMIC);
1001                         if (err)
1002                                 return err;
1003                         add_helper = true;
1004
1005                         /* helper installed, add seqadj if NAT is required */
1006                         if (info->nat && !nfct_seqadj(ct)) {
1007                                 if (!nfct_seqadj_ext_add(ct))
1008                                         return -EINVAL;
1009                         }
1010                 }
1011
1012                 /* Call the helper only if:
1013                  * - nf_conntrack_in() was executed above ("!cached") or a
1014                  *   helper was just attached ("add_helper") for a confirmed
1015                  *   connection, or
1016                  * - When committing an unconfirmed connection.
1017                  */
1018                 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1019                                               info->commit) &&
1020                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1021                         return -EINVAL;
1022                 }
1023         }
1024
1025         return 0;
1026 }
1027
1028 /* Lookup connection and read fields into key. */
1029 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1030                          const struct ovs_conntrack_info *info,
1031                          struct sk_buff *skb)
1032 {
1033         struct nf_conntrack_expect *exp;
1034
1035         /* If we pass an expected packet through nf_conntrack_in() the
1036          * expectation is typically removed, but the packet could still be
1037          * lost in upcall processing.  To prevent this from happening we
1038          * perform an explicit expectation lookup.  Expected connections are
1039          * always new, and will be passed through conntrack only when they are
1040          * committed, as it is OK to remove the expectation at that time.
1041          */
1042         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1043         if (exp) {
1044                 u8 state;
1045
1046                 /* NOTE: New connections are NATted and Helped only when
1047                  * committed, so we are not calling into NAT here.
1048                  */
1049                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1050                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1051         } else {
1052                 struct nf_conn *ct;
1053                 int err;
1054
1055                 err = __ovs_ct_lookup(net, key, info, skb);
1056                 if (err)
1057                         return err;
1058
1059                 ct = (struct nf_conn *)skb_nfct(skb);
1060                 if (ct)
1061                         nf_ct_deliver_cached_events(ct);
1062         }
1063
1064         return 0;
1065 }
1066
1067 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1068 {
1069         size_t i;
1070
1071         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1072                 if (labels->ct_labels_32[i])
1073                         return true;
1074
1075         return false;
1076 }
1077
1078 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1079 static struct hlist_head *ct_limit_hash_bucket(
1080         const struct ovs_ct_limit_info *info, u16 zone)
1081 {
1082         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1083 }
1084
1085 /* Call with ovs_mutex */
1086 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1087                          struct ovs_ct_limit *new_ct_limit)
1088 {
1089         struct ovs_ct_limit *ct_limit;
1090         struct hlist_head *head;
1091
1092         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1093         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1094                 if (ct_limit->zone == new_ct_limit->zone) {
1095                         hlist_replace_rcu(&ct_limit->hlist_node,
1096                                           &new_ct_limit->hlist_node);
1097                         kfree_rcu(ct_limit, rcu);
1098                         return;
1099                 }
1100         }
1101
1102         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1103 }
1104
1105 /* Call with ovs_mutex */
1106 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1107 {
1108         struct ovs_ct_limit *ct_limit;
1109         struct hlist_head *head;
1110         struct hlist_node *n;
1111
1112         head = ct_limit_hash_bucket(info, zone);
1113         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1114                 if (ct_limit->zone == zone) {
1115                         hlist_del_rcu(&ct_limit->hlist_node);
1116                         kfree_rcu(ct_limit, rcu);
1117                         return;
1118                 }
1119         }
1120 }
1121
1122 /* Call with RCU read lock */
1123 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1124 {
1125         struct ovs_ct_limit *ct_limit;
1126         struct hlist_head *head;
1127
1128         head = ct_limit_hash_bucket(info, zone);
1129         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1130                 if (ct_limit->zone == zone)
1131                         return ct_limit->limit;
1132         }
1133
1134         return info->default_limit;
1135 }
1136
1137 static int ovs_ct_check_limit(struct net *net,
1138                               const struct ovs_conntrack_info *info,
1139                               const struct nf_conntrack_tuple *tuple)
1140 {
1141         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1142         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1143         u32 per_zone_limit, connections;
1144         u32 conncount_key;
1145
1146         conncount_key = info->zone.id;
1147
1148         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1149         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1150                 return 0;
1151
1152         connections = nf_conncount_count(net, ct_limit_info->data,
1153                                          &conncount_key, tuple, &info->zone);
1154         if (connections > per_zone_limit)
1155                 return -ENOMEM;
1156
1157         return 0;
1158 }
1159 #endif
1160
1161 /* Lookup connection and confirm if unconfirmed. */
1162 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1163                          const struct ovs_conntrack_info *info,
1164                          struct sk_buff *skb)
1165 {
1166         enum ip_conntrack_info ctinfo;
1167         struct nf_conn *ct;
1168         int err;
1169
1170         err = __ovs_ct_lookup(net, key, info, skb);
1171         if (err)
1172                 return err;
1173
1174         /* The connection could be invalid, in which case this is a no-op.*/
1175         ct = nf_ct_get(skb, &ctinfo);
1176         if (!ct)
1177                 return 0;
1178
1179 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1180         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1181                 if (!nf_ct_is_confirmed(ct)) {
1182                         err = ovs_ct_check_limit(net, info,
1183                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1184                         if (err) {
1185                                 net_warn_ratelimited("openvswitch: zone: %u "
1186                                         "exceeds conntrack limit\n",
1187                                         info->zone.id);
1188                                 return err;
1189                         }
1190                 }
1191         }
1192 #endif
1193
1194         /* Set the conntrack event mask if given.  NEW and DELETE events have
1195          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1196          * typically would receive many kinds of updates.  Setting the event
1197          * mask allows those events to be filtered.  The set event mask will
1198          * remain in effect for the lifetime of the connection unless changed
1199          * by a further CT action with both the commit flag and the eventmask
1200          * option. */
1201         if (info->have_eventmask) {
1202                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1203
1204                 if (cache)
1205                         cache->ctmask = info->eventmask;
1206         }
1207
1208         /* Apply changes before confirming the connection so that the initial
1209          * conntrack NEW netlink event carries the values given in the CT
1210          * action.
1211          */
1212         if (info->mark.mask) {
1213                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1214                                       info->mark.mask);
1215                 if (err)
1216                         return err;
1217         }
1218         if (!nf_ct_is_confirmed(ct)) {
1219                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1220                                          &info->labels.mask);
1221                 if (err)
1222                         return err;
1223         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1224                    labels_nonzero(&info->labels.mask)) {
1225                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1226                                         &info->labels.mask);
1227                 if (err)
1228                         return err;
1229         }
1230         /* This will take care of sending queued events even if the connection
1231          * is already confirmed.
1232          */
1233         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1234                 return -EINVAL;
1235
1236         return 0;
1237 }
1238
1239 /* Trim the skb to the length specified by the IP/IPv6 header,
1240  * removing any trailing lower-layer padding. This prepares the skb
1241  * for higher-layer processing that assumes skb->len excludes padding
1242  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1243  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1244  */
1245 static int ovs_skb_network_trim(struct sk_buff *skb)
1246 {
1247         unsigned int len;
1248         int err;
1249
1250         switch (skb->protocol) {
1251         case htons(ETH_P_IP):
1252                 len = ntohs(ip_hdr(skb)->tot_len);
1253                 break;
1254         case htons(ETH_P_IPV6):
1255                 len = sizeof(struct ipv6hdr)
1256                         + ntohs(ipv6_hdr(skb)->payload_len);
1257                 break;
1258         default:
1259                 len = skb->len;
1260         }
1261
1262         err = pskb_trim_rcsum(skb, len);
1263         if (err)
1264                 kfree_skb(skb);
1265
1266         return err;
1267 }
1268
1269 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1270  * value if 'skb' is freed.
1271  */
1272 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1273                    struct sw_flow_key *key,
1274                    const struct ovs_conntrack_info *info)
1275 {
1276         int nh_ofs;
1277         int err;
1278
1279         /* The conntrack module expects to be working at L3. */
1280         nh_ofs = skb_network_offset(skb);
1281         skb_pull_rcsum(skb, nh_ofs);
1282
1283         err = ovs_skb_network_trim(skb);
1284         if (err)
1285                 return err;
1286
1287         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1288                 err = handle_fragments(net, key, info->zone.id, skb);
1289                 if (err)
1290                         return err;
1291         }
1292
1293         if (info->commit)
1294                 err = ovs_ct_commit(net, key, info, skb);
1295         else
1296                 err = ovs_ct_lookup(net, key, info, skb);
1297
1298         skb_push(skb, nh_ofs);
1299         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1300         if (err)
1301                 kfree_skb(skb);
1302         return err;
1303 }
1304
1305 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1306 {
1307         if (skb_nfct(skb)) {
1308                 nf_conntrack_put(skb_nfct(skb));
1309                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1310                 ovs_ct_fill_key(skb, key);
1311         }
1312
1313         return 0;
1314 }
1315
1316 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1317                              const struct sw_flow_key *key, bool log)
1318 {
1319         struct nf_conntrack_helper *helper;
1320         struct nf_conn_help *help;
1321         int ret = 0;
1322
1323         helper = nf_conntrack_helper_try_module_get(name, info->family,
1324                                                     key->ip.proto);
1325         if (!helper) {
1326                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1327                 return -EINVAL;
1328         }
1329
1330         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1331         if (!help) {
1332                 nf_conntrack_helper_put(helper);
1333                 return -ENOMEM;
1334         }
1335
1336 #if IS_ENABLED(CONFIG_NF_NAT)
1337         if (info->nat) {
1338                 ret = nf_nat_helper_try_module_get(name, info->family,
1339                                                    key->ip.proto);
1340                 if (ret) {
1341                         nf_conntrack_helper_put(helper);
1342                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1343                                   name, ret);
1344                         return ret;
1345                 }
1346         }
1347 #endif
1348         rcu_assign_pointer(help->helper, helper);
1349         info->helper = helper;
1350         return ret;
1351 }
1352
1353 #if IS_ENABLED(CONFIG_NF_NAT)
1354 static int parse_nat(const struct nlattr *attr,
1355                      struct ovs_conntrack_info *info, bool log)
1356 {
1357         struct nlattr *a;
1358         int rem;
1359         bool have_ip_max = false;
1360         bool have_proto_max = false;
1361         bool ip_vers = (info->family == NFPROTO_IPV6);
1362
1363         nla_for_each_nested(a, attr, rem) {
1364                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1365                         [OVS_NAT_ATTR_SRC] = {0, 0},
1366                         [OVS_NAT_ATTR_DST] = {0, 0},
1367                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1368                                                  sizeof(struct in6_addr)},
1369                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1370                                                  sizeof(struct in6_addr)},
1371                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1372                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1373                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1374                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1375                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1376                 };
1377                 int type = nla_type(a);
1378
1379                 if (type > OVS_NAT_ATTR_MAX) {
1380                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1381                                   type, OVS_NAT_ATTR_MAX);
1382                         return -EINVAL;
1383                 }
1384
1385                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1386                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1387                                   type, nla_len(a),
1388                                   ovs_nat_attr_lens[type][ip_vers]);
1389                         return -EINVAL;
1390                 }
1391
1392                 switch (type) {
1393                 case OVS_NAT_ATTR_SRC:
1394                 case OVS_NAT_ATTR_DST:
1395                         if (info->nat) {
1396                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1397                                 return -ERANGE;
1398                         }
1399                         info->nat |= OVS_CT_NAT;
1400                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1401                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1402                         break;
1403
1404                 case OVS_NAT_ATTR_IP_MIN:
1405                         nla_memcpy(&info->range.min_addr, a,
1406                                    sizeof(info->range.min_addr));
1407                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1408                         break;
1409
1410                 case OVS_NAT_ATTR_IP_MAX:
1411                         have_ip_max = true;
1412                         nla_memcpy(&info->range.max_addr, a,
1413                                    sizeof(info->range.max_addr));
1414                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1415                         break;
1416
1417                 case OVS_NAT_ATTR_PROTO_MIN:
1418                         info->range.min_proto.all = htons(nla_get_u16(a));
1419                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1420                         break;
1421
1422                 case OVS_NAT_ATTR_PROTO_MAX:
1423                         have_proto_max = true;
1424                         info->range.max_proto.all = htons(nla_get_u16(a));
1425                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1426                         break;
1427
1428                 case OVS_NAT_ATTR_PERSISTENT:
1429                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1430                         break;
1431
1432                 case OVS_NAT_ATTR_PROTO_HASH:
1433                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1434                         break;
1435
1436                 case OVS_NAT_ATTR_PROTO_RANDOM:
1437                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1438                         break;
1439
1440                 default:
1441                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1442                         return -EINVAL;
1443                 }
1444         }
1445
1446         if (rem > 0) {
1447                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1448                 return -EINVAL;
1449         }
1450         if (!info->nat) {
1451                 /* Do not allow flags if no type is given. */
1452                 if (info->range.flags) {
1453                         OVS_NLERR(log,
1454                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1455                                   );
1456                         return -EINVAL;
1457                 }
1458                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1459         } else if (!info->commit) {
1460                 OVS_NLERR(log,
1461                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1462                           );
1463                 return -EINVAL;
1464         }
1465         /* Allow missing IP_MAX. */
1466         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1467                 memcpy(&info->range.max_addr, &info->range.min_addr,
1468                        sizeof(info->range.max_addr));
1469         }
1470         /* Allow missing PROTO_MAX. */
1471         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1472             !have_proto_max) {
1473                 info->range.max_proto.all = info->range.min_proto.all;
1474         }
1475         return 0;
1476 }
1477 #endif
1478
1479 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1480         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1481         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1482         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1483                                     .maxlen = sizeof(u16) },
1484         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1485                                     .maxlen = sizeof(struct md_mark) },
1486         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1487                                     .maxlen = sizeof(struct md_labels) },
1488         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1489                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1490 #if IS_ENABLED(CONFIG_NF_NAT)
1491         /* NAT length is checked when parsing the nested attributes. */
1492         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1493 #endif
1494         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1495                                     .maxlen = sizeof(u32) },
1496         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1497                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1498 };
1499
1500 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1501                     const char **helper, bool log)
1502 {
1503         struct nlattr *a;
1504         int rem;
1505
1506         nla_for_each_nested(a, attr, rem) {
1507                 int type = nla_type(a);
1508                 int maxlen;
1509                 int minlen;
1510
1511                 if (type > OVS_CT_ATTR_MAX) {
1512                         OVS_NLERR(log,
1513                                   "Unknown conntrack attr (type=%d, max=%d)",
1514                                   type, OVS_CT_ATTR_MAX);
1515                         return -EINVAL;
1516                 }
1517
1518                 maxlen = ovs_ct_attr_lens[type].maxlen;
1519                 minlen = ovs_ct_attr_lens[type].minlen;
1520                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1521                         OVS_NLERR(log,
1522                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1523                                   type, nla_len(a), maxlen);
1524                         return -EINVAL;
1525                 }
1526
1527                 switch (type) {
1528                 case OVS_CT_ATTR_FORCE_COMMIT:
1529                         info->force = true;
1530                         /* fall through. */
1531                 case OVS_CT_ATTR_COMMIT:
1532                         info->commit = true;
1533                         break;
1534 #ifdef CONFIG_NF_CONNTRACK_ZONES
1535                 case OVS_CT_ATTR_ZONE:
1536                         info->zone.id = nla_get_u16(a);
1537                         break;
1538 #endif
1539 #ifdef CONFIG_NF_CONNTRACK_MARK
1540                 case OVS_CT_ATTR_MARK: {
1541                         struct md_mark *mark = nla_data(a);
1542
1543                         if (!mark->mask) {
1544                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1545                                 return -EINVAL;
1546                         }
1547                         info->mark = *mark;
1548                         break;
1549                 }
1550 #endif
1551 #ifdef CONFIG_NF_CONNTRACK_LABELS
1552                 case OVS_CT_ATTR_LABELS: {
1553                         struct md_labels *labels = nla_data(a);
1554
1555                         if (!labels_nonzero(&labels->mask)) {
1556                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1557                                 return -EINVAL;
1558                         }
1559                         info->labels = *labels;
1560                         break;
1561                 }
1562 #endif
1563                 case OVS_CT_ATTR_HELPER:
1564                         *helper = nla_data(a);
1565                         if (!memchr(*helper, '\0', nla_len(a))) {
1566                                 OVS_NLERR(log, "Invalid conntrack helper");
1567                                 return -EINVAL;
1568                         }
1569                         break;
1570 #if IS_ENABLED(CONFIG_NF_NAT)
1571                 case OVS_CT_ATTR_NAT: {
1572                         int err = parse_nat(a, info, log);
1573
1574                         if (err)
1575                                 return err;
1576                         break;
1577                 }
1578 #endif
1579                 case OVS_CT_ATTR_EVENTMASK:
1580                         info->have_eventmask = true;
1581                         info->eventmask = nla_get_u32(a);
1582                         break;
1583 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1584                 case OVS_CT_ATTR_TIMEOUT:
1585                         memcpy(info->timeout, nla_data(a), nla_len(a));
1586                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1587                                 OVS_NLERR(log, "Invalid conntrack timeout");
1588                                 return -EINVAL;
1589                         }
1590                         break;
1591 #endif
1592
1593                 default:
1594                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1595                                   type);
1596                         return -EINVAL;
1597                 }
1598         }
1599
1600 #ifdef CONFIG_NF_CONNTRACK_MARK
1601         if (!info->commit && info->mark.mask) {
1602                 OVS_NLERR(log,
1603                           "Setting conntrack mark requires 'commit' flag.");
1604                 return -EINVAL;
1605         }
1606 #endif
1607 #ifdef CONFIG_NF_CONNTRACK_LABELS
1608         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1609                 OVS_NLERR(log,
1610                           "Setting conntrack labels requires 'commit' flag.");
1611                 return -EINVAL;
1612         }
1613 #endif
1614         if (rem > 0) {
1615                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1616                 return -EINVAL;
1617         }
1618
1619         return 0;
1620 }
1621
1622 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1623 {
1624         if (attr == OVS_KEY_ATTR_CT_STATE)
1625                 return true;
1626         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1627             attr == OVS_KEY_ATTR_CT_ZONE)
1628                 return true;
1629         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1630             attr == OVS_KEY_ATTR_CT_MARK)
1631                 return true;
1632         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1633             attr == OVS_KEY_ATTR_CT_LABELS) {
1634                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1635
1636                 return ovs_net->xt_label;
1637         }
1638
1639         return false;
1640 }
1641
1642 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1643                        const struct sw_flow_key *key,
1644                        struct sw_flow_actions **sfa,  bool log)
1645 {
1646         struct ovs_conntrack_info ct_info;
1647         const char *helper = NULL;
1648         u16 family;
1649         int err;
1650
1651         family = key_to_nfproto(key);
1652         if (family == NFPROTO_UNSPEC) {
1653                 OVS_NLERR(log, "ct family unspecified");
1654                 return -EINVAL;
1655         }
1656
1657         memset(&ct_info, 0, sizeof(ct_info));
1658         ct_info.family = family;
1659
1660         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1661                         NF_CT_DEFAULT_ZONE_DIR, 0);
1662
1663         err = parse_ct(attr, &ct_info, &helper, log);
1664         if (err)
1665                 return err;
1666
1667         /* Set up template for tracking connections in specific zones. */
1668         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1669         if (!ct_info.ct) {
1670                 OVS_NLERR(log, "Failed to allocate conntrack template");
1671                 return -ENOMEM;
1672         }
1673
1674         if (ct_info.timeout[0]) {
1675                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1676                                       ct_info.timeout))
1677                         pr_info_ratelimited("Failed to associated timeout "
1678                                             "policy `%s'\n", ct_info.timeout);
1679                 else
1680                         ct_info.nf_ct_timeout = rcu_dereference(
1681                                 nf_ct_timeout_find(ct_info.ct)->timeout);
1682
1683         }
1684
1685         if (helper) {
1686                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1687                 if (err)
1688                         goto err_free_ct;
1689         }
1690
1691         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1692                                  sizeof(ct_info), log);
1693         if (err)
1694                 goto err_free_ct;
1695
1696         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1697         nf_conntrack_get(&ct_info.ct->ct_general);
1698         return 0;
1699 err_free_ct:
1700         __ovs_ct_free_action(&ct_info);
1701         return err;
1702 }
1703
1704 #if IS_ENABLED(CONFIG_NF_NAT)
1705 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1706                                struct sk_buff *skb)
1707 {
1708         struct nlattr *start;
1709
1710         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1711         if (!start)
1712                 return false;
1713
1714         if (info->nat & OVS_CT_SRC_NAT) {
1715                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1716                         return false;
1717         } else if (info->nat & OVS_CT_DST_NAT) {
1718                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1719                         return false;
1720         } else {
1721                 goto out;
1722         }
1723
1724         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1725                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1726                     info->family == NFPROTO_IPV4) {
1727                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1728                                             info->range.min_addr.ip) ||
1729                             (info->range.max_addr.ip
1730                              != info->range.min_addr.ip &&
1731                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1732                                               info->range.max_addr.ip))))
1733                                 return false;
1734                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1735                            info->family == NFPROTO_IPV6) {
1736                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1737                                              &info->range.min_addr.in6) ||
1738                             (memcmp(&info->range.max_addr.in6,
1739                                     &info->range.min_addr.in6,
1740                                     sizeof(info->range.max_addr.in6)) &&
1741                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1742                                                &info->range.max_addr.in6))))
1743                                 return false;
1744                 } else {
1745                         return false;
1746                 }
1747         }
1748         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1749             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1750                          ntohs(info->range.min_proto.all)) ||
1751              (info->range.max_proto.all != info->range.min_proto.all &&
1752               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1753                           ntohs(info->range.max_proto.all)))))
1754                 return false;
1755
1756         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1757             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1758                 return false;
1759         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1760             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1761                 return false;
1762         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1763             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1764                 return false;
1765 out:
1766         nla_nest_end(skb, start);
1767
1768         return true;
1769 }
1770 #endif
1771
1772 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1773                           struct sk_buff *skb)
1774 {
1775         struct nlattr *start;
1776
1777         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1778         if (!start)
1779                 return -EMSGSIZE;
1780
1781         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1782                                             ? OVS_CT_ATTR_FORCE_COMMIT
1783                                             : OVS_CT_ATTR_COMMIT))
1784                 return -EMSGSIZE;
1785         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1786             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1787                 return -EMSGSIZE;
1788         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1789             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1790                     &ct_info->mark))
1791                 return -EMSGSIZE;
1792         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1793             labels_nonzero(&ct_info->labels.mask) &&
1794             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1795                     &ct_info->labels))
1796                 return -EMSGSIZE;
1797         if (ct_info->helper) {
1798                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1799                                    ct_info->helper->name))
1800                         return -EMSGSIZE;
1801         }
1802         if (ct_info->have_eventmask &&
1803             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1804                 return -EMSGSIZE;
1805         if (ct_info->timeout[0]) {
1806                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1807                         return -EMSGSIZE;
1808         }
1809
1810 #if IS_ENABLED(CONFIG_NF_NAT)
1811         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1812                 return -EMSGSIZE;
1813 #endif
1814         nla_nest_end(skb, start);
1815
1816         return 0;
1817 }
1818
1819 void ovs_ct_free_action(const struct nlattr *a)
1820 {
1821         struct ovs_conntrack_info *ct_info = nla_data(a);
1822
1823         __ovs_ct_free_action(ct_info);
1824 }
1825
1826 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1827 {
1828         if (ct_info->helper) {
1829 #if IS_ENABLED(CONFIG_NF_NAT)
1830                 if (ct_info->nat)
1831                         nf_nat_helper_put(ct_info->helper);
1832 #endif
1833                 nf_conntrack_helper_put(ct_info->helper);
1834         }
1835         if (ct_info->ct) {
1836                 if (ct_info->timeout[0])
1837                         nf_ct_destroy_timeout(ct_info->ct);
1838                 nf_ct_tmpl_free(ct_info->ct);
1839         }
1840 }
1841
1842 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1843 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1844 {
1845         int i, err;
1846
1847         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1848                                          GFP_KERNEL);
1849         if (!ovs_net->ct_limit_info)
1850                 return -ENOMEM;
1851
1852         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1853         ovs_net->ct_limit_info->limits =
1854                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1855                               GFP_KERNEL);
1856         if (!ovs_net->ct_limit_info->limits) {
1857                 kfree(ovs_net->ct_limit_info);
1858                 return -ENOMEM;
1859         }
1860
1861         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1862                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1863
1864         ovs_net->ct_limit_info->data =
1865                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1866
1867         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1868                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1869                 kfree(ovs_net->ct_limit_info->limits);
1870                 kfree(ovs_net->ct_limit_info);
1871                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1872                 return err;
1873         }
1874         return 0;
1875 }
1876
1877 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1878 {
1879         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1880         int i;
1881
1882         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1883         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1884                 struct hlist_head *head = &info->limits[i];
1885                 struct ovs_ct_limit *ct_limit;
1886
1887                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1888                         kfree_rcu(ct_limit, rcu);
1889         }
1890         kfree(ovs_net->ct_limit_info->limits);
1891         kfree(ovs_net->ct_limit_info);
1892 }
1893
1894 static struct sk_buff *
1895 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1896                              struct ovs_header **ovs_reply_header)
1897 {
1898         struct ovs_header *ovs_header = info->userhdr;
1899         struct sk_buff *skb;
1900
1901         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1902         if (!skb)
1903                 return ERR_PTR(-ENOMEM);
1904
1905         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1906                                         info->snd_seq,
1907                                         &dp_ct_limit_genl_family, 0, cmd);
1908
1909         if (!*ovs_reply_header) {
1910                 nlmsg_free(skb);
1911                 return ERR_PTR(-EMSGSIZE);
1912         }
1913         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1914
1915         return skb;
1916 }
1917
1918 static bool check_zone_id(int zone_id, u16 *pzone)
1919 {
1920         if (zone_id >= 0 && zone_id <= 65535) {
1921                 *pzone = (u16)zone_id;
1922                 return true;
1923         }
1924         return false;
1925 }
1926
1927 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1928                                        struct ovs_ct_limit_info *info)
1929 {
1930         struct ovs_zone_limit *zone_limit;
1931         int rem;
1932         u16 zone;
1933
1934         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1935         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1936
1937         while (rem >= sizeof(*zone_limit)) {
1938                 if (unlikely(zone_limit->zone_id ==
1939                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1940                         ovs_lock();
1941                         info->default_limit = zone_limit->limit;
1942                         ovs_unlock();
1943                 } else if (unlikely(!check_zone_id(
1944                                 zone_limit->zone_id, &zone))) {
1945                         OVS_NLERR(true, "zone id is out of range");
1946                 } else {
1947                         struct ovs_ct_limit *ct_limit;
1948
1949                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1950                         if (!ct_limit)
1951                                 return -ENOMEM;
1952
1953                         ct_limit->zone = zone;
1954                         ct_limit->limit = zone_limit->limit;
1955
1956                         ovs_lock();
1957                         ct_limit_set(info, ct_limit);
1958                         ovs_unlock();
1959                 }
1960                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1961                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1962                                 NLA_ALIGN(sizeof(*zone_limit)));
1963         }
1964
1965         if (rem)
1966                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1967
1968         return 0;
1969 }
1970
1971 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1972                                        struct ovs_ct_limit_info *info)
1973 {
1974         struct ovs_zone_limit *zone_limit;
1975         int rem;
1976         u16 zone;
1977
1978         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1979         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1980
1981         while (rem >= sizeof(*zone_limit)) {
1982                 if (unlikely(zone_limit->zone_id ==
1983                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1984                         ovs_lock();
1985                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1986                         ovs_unlock();
1987                 } else if (unlikely(!check_zone_id(
1988                                 zone_limit->zone_id, &zone))) {
1989                         OVS_NLERR(true, "zone id is out of range");
1990                 } else {
1991                         ovs_lock();
1992                         ct_limit_del(info, zone);
1993                         ovs_unlock();
1994                 }
1995                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1996                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1997                                 NLA_ALIGN(sizeof(*zone_limit)));
1998         }
1999
2000         if (rem)
2001                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2002
2003         return 0;
2004 }
2005
2006 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2007                                           struct sk_buff *reply)
2008 {
2009         struct ovs_zone_limit zone_limit;
2010         int err;
2011
2012         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2013         zone_limit.limit = info->default_limit;
2014         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2015         if (err)
2016                 return err;
2017
2018         return 0;
2019 }
2020
2021 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2022                                          struct nf_conncount_data *data,
2023                                          u16 zone_id, u32 limit,
2024                                          struct sk_buff *reply)
2025 {
2026         struct nf_conntrack_zone ct_zone;
2027         struct ovs_zone_limit zone_limit;
2028         u32 conncount_key = zone_id;
2029
2030         zone_limit.zone_id = zone_id;
2031         zone_limit.limit = limit;
2032         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2033
2034         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2035                                               &ct_zone);
2036         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2037 }
2038
2039 static int ovs_ct_limit_get_zone_limit(struct net *net,
2040                                        struct nlattr *nla_zone_limit,
2041                                        struct ovs_ct_limit_info *info,
2042                                        struct sk_buff *reply)
2043 {
2044         struct ovs_zone_limit *zone_limit;
2045         int rem, err;
2046         u32 limit;
2047         u16 zone;
2048
2049         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2050         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2051
2052         while (rem >= sizeof(*zone_limit)) {
2053                 if (unlikely(zone_limit->zone_id ==
2054                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2055                         err = ovs_ct_limit_get_default_limit(info, reply);
2056                         if (err)
2057                                 return err;
2058                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2059                                                         &zone))) {
2060                         OVS_NLERR(true, "zone id is out of range");
2061                 } else {
2062                         rcu_read_lock();
2063                         limit = ct_limit_get(info, zone);
2064                         rcu_read_unlock();
2065
2066                         err = __ovs_ct_limit_get_zone_limit(
2067                                 net, info->data, zone, limit, reply);
2068                         if (err)
2069                                 return err;
2070                 }
2071                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2072                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2073                                 NLA_ALIGN(sizeof(*zone_limit)));
2074         }
2075
2076         if (rem)
2077                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2078
2079         return 0;
2080 }
2081
2082 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2083                                            struct ovs_ct_limit_info *info,
2084                                            struct sk_buff *reply)
2085 {
2086         struct ovs_ct_limit *ct_limit;
2087         struct hlist_head *head;
2088         int i, err = 0;
2089
2090         err = ovs_ct_limit_get_default_limit(info, reply);
2091         if (err)
2092                 return err;
2093
2094         rcu_read_lock();
2095         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2096                 head = &info->limits[i];
2097                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2098                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2099                                 ct_limit->zone, ct_limit->limit, reply);
2100                         if (err)
2101                                 goto exit_err;
2102                 }
2103         }
2104
2105 exit_err:
2106         rcu_read_unlock();
2107         return err;
2108 }
2109
2110 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2111 {
2112         struct nlattr **a = info->attrs;
2113         struct sk_buff *reply;
2114         struct ovs_header *ovs_reply_header;
2115         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2116         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2117         int err;
2118
2119         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2120                                              &ovs_reply_header);
2121         if (IS_ERR(reply))
2122                 return PTR_ERR(reply);
2123
2124         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2125                 err = -EINVAL;
2126                 goto exit_err;
2127         }
2128
2129         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2130                                           ct_limit_info);
2131         if (err)
2132                 goto exit_err;
2133
2134         static_branch_enable(&ovs_ct_limit_enabled);
2135
2136         genlmsg_end(reply, ovs_reply_header);
2137         return genlmsg_reply(reply, info);
2138
2139 exit_err:
2140         nlmsg_free(reply);
2141         return err;
2142 }
2143
2144 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2145 {
2146         struct nlattr **a = info->attrs;
2147         struct sk_buff *reply;
2148         struct ovs_header *ovs_reply_header;
2149         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2150         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2151         int err;
2152
2153         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2154                                              &ovs_reply_header);
2155         if (IS_ERR(reply))
2156                 return PTR_ERR(reply);
2157
2158         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2159                 err = -EINVAL;
2160                 goto exit_err;
2161         }
2162
2163         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2164                                           ct_limit_info);
2165         if (err)
2166                 goto exit_err;
2167
2168         genlmsg_end(reply, ovs_reply_header);
2169         return genlmsg_reply(reply, info);
2170
2171 exit_err:
2172         nlmsg_free(reply);
2173         return err;
2174 }
2175
2176 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2177 {
2178         struct nlattr **a = info->attrs;
2179         struct nlattr *nla_reply;
2180         struct sk_buff *reply;
2181         struct ovs_header *ovs_reply_header;
2182         struct net *net = sock_net(skb->sk);
2183         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2184         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2185         int err;
2186
2187         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2188                                              &ovs_reply_header);
2189         if (IS_ERR(reply))
2190                 return PTR_ERR(reply);
2191
2192         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2193         if (!nla_reply) {
2194                 err = -EMSGSIZE;
2195                 goto exit_err;
2196         }
2197
2198         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2199                 err = ovs_ct_limit_get_zone_limit(
2200                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2201                         reply);
2202                 if (err)
2203                         goto exit_err;
2204         } else {
2205                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2206                                                       reply);
2207                 if (err)
2208                         goto exit_err;
2209         }
2210
2211         nla_nest_end(reply, nla_reply);
2212         genlmsg_end(reply, ovs_reply_header);
2213         return genlmsg_reply(reply, info);
2214
2215 exit_err:
2216         nlmsg_free(reply);
2217         return err;
2218 }
2219
2220 static struct genl_ops ct_limit_genl_ops[] = {
2221         { .cmd = OVS_CT_LIMIT_CMD_SET,
2222                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2223                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2224                                            * privilege. */
2225                 .doit = ovs_ct_limit_cmd_set,
2226         },
2227         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2228                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2229                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2230                                            * privilege. */
2231                 .doit = ovs_ct_limit_cmd_del,
2232         },
2233         { .cmd = OVS_CT_LIMIT_CMD_GET,
2234                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2235                 .flags = 0,               /* OK for unprivileged users. */
2236                 .doit = ovs_ct_limit_cmd_get,
2237         },
2238 };
2239
2240 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2241         .name = OVS_CT_LIMIT_MCGROUP,
2242 };
2243
2244 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2245         .hdrsize = sizeof(struct ovs_header),
2246         .name = OVS_CT_LIMIT_FAMILY,
2247         .version = OVS_CT_LIMIT_VERSION,
2248         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2249         .policy = ct_limit_policy,
2250         .netnsok = true,
2251         .parallel_ops = true,
2252         .ops = ct_limit_genl_ops,
2253         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2254         .mcgrps = &ovs_ct_limit_multicast_group,
2255         .n_mcgrps = 1,
2256         .module = THIS_MODULE,
2257 };
2258 #endif
2259
2260 int ovs_ct_init(struct net *net)
2261 {
2262         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2263         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2264
2265         if (nf_connlabels_get(net, n_bits - 1)) {
2266                 ovs_net->xt_label = false;
2267                 OVS_NLERR(true, "Failed to set connlabel length");
2268         } else {
2269                 ovs_net->xt_label = true;
2270         }
2271
2272 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2273         return ovs_ct_limit_init(net, ovs_net);
2274 #else
2275         return 0;
2276 #endif
2277 }
2278
2279 void ovs_ct_exit(struct net *net)
2280 {
2281         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2282
2283 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2284         ovs_ct_limit_exit(net, ovs_net);
2285 #endif
2286
2287         if (ovs_net->xt_label)
2288                 nf_connlabels_put(net);
2289 }