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 #if IS_ENABLED(CONFIG_NF_NAT)
  71         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  72 #endif
  73 };
  74
  75 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  76 #define OVS_CT_LIMIT_UNLIMITED  0
  77 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  78 #define CT_LIMIT_HASH_BUCKETS 512
  79 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  80
  81 struct ovs_ct_limit {
  82         /* Elements in ovs_ct_limit_info->limits hash table */
  83         struct hlist_node hlist_node;
  84         struct rcu_head rcu;
  85         u16 zone;
  86         u32 limit;
  87 };
  88
  89 struct ovs_ct_limit_info {
  90         u32 default_limit;
  91         struct hlist_head *limits;
  92         struct nf_conncount_data *data;
  93 };
  94
  95 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
  96         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
  97 };
  98 #endif
  99
 100 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 101
 102 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 103
 104 static u16 key_to_nfproto(const struct sw_flow_key *key)
 105 {
 106         switch (ntohs(key->eth.type)) {
 107         case ETH_P_IP:
 108                 return NFPROTO_IPV4;
 109         case ETH_P_IPV6:
 110                 return NFPROTO_IPV6;
 111         default:
 112                 return NFPROTO_UNSPEC;
 113         }
 114 }
 115
 116 /* Map SKB connection state into the values used by flow definition. */
 117 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 118 {
 119         u8 ct_state = OVS_CS_F_TRACKED;
 120
 121         switch (ctinfo) {
 122         case IP_CT_ESTABLISHED_REPLY:
 123         case IP_CT_RELATED_REPLY:
 124                 ct_state |= OVS_CS_F_REPLY_DIR;
 125                 break;
 126         default:
 127                 break;
 128         }
 129
 130         switch (ctinfo) {
 131         case IP_CT_ESTABLISHED:
 132         case IP_CT_ESTABLISHED_REPLY:
 133                 ct_state |= OVS_CS_F_ESTABLISHED;
 134                 break;
 135         case IP_CT_RELATED:
 136         case IP_CT_RELATED_REPLY:
 137                 ct_state |= OVS_CS_F_RELATED;
 138                 break;
 139         case IP_CT_NEW:
 140                 ct_state |= OVS_CS_F_NEW;
 141                 break;
 142         default:
 143                 break;
 144         }
 145
 146         return ct_state;
 147 }
 148
 149 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 150 {
 151 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 152         return ct ? ct->mark : 0;
 153 #else
 154         return 0;
 155 #endif
 156 }
 157
 158 /* Guard against conntrack labels max size shrinking below 128 bits. */
 159 #if NF_CT_LABELS_MAX_SIZE < 16
 160 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 161 #endif
 162
 163 static void ovs_ct_get_labels(const struct nf_conn *ct,
 164                               struct ovs_key_ct_labels *labels)
 165 {
 166         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 167
 168         if (cl)
 169                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 170         else
 171                 memset(labels, 0, OVS_CT_LABELS_LEN);
 172 }
 173
 174 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 175                                         const struct nf_conntrack_tuple *orig,
 176                                         u8 icmp_proto)
 177 {
 178         key->ct_orig_proto = orig->dst.protonum;
 179         if (orig->dst.protonum == icmp_proto) {
 180                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 181                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 182         } else {
 183                 key->ct.orig_tp.src = orig->src.u.all;
 184                 key->ct.orig_tp.dst = orig->dst.u.all;
 185         }
 186 }
 187
 188 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 189                                 const struct nf_conntrack_zone *zone,
 190                                 const struct nf_conn *ct)
 191 {
 192         key->ct_state = state;
 193         key->ct_zone = zone->id;
 194         key->ct.mark = ovs_ct_get_mark(ct);
 195         ovs_ct_get_labels(ct, &key->ct.labels);
 196
 197         if (ct) {
 198                 const struct nf_conntrack_tuple *orig;
 199
 200                 /* Use the master if we have one. */
 201                 if (ct->master)
 202                         ct = ct->master;
 203                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 204
 205                 /* IP version must match with the master connection. */
 206                 if (key->eth.type == htons(ETH_P_IP) &&
 207                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 208                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 209                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 210                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 211                         return;
 212                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 213                            !sw_flow_key_is_nd(key) &&
 214                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 215                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 216                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 217                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 218                         return;
 219                 }
 220         }
 221         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 222          * original direction key fields.
 223          */
 224         key->ct_orig_proto = 0;
 225 }
 226
 227 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 228  * previously sent the packet to conntrack via the ct action.  If
 229  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 230  * initialized from the connection status.
 231  */
 232 static void ovs_ct_update_key(const struct sk_buff *skb,
 233                               const struct ovs_conntrack_info *info,
 234                               struct sw_flow_key *key, bool post_ct,
 235                               bool keep_nat_flags)
 236 {
 237         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 238         enum ip_conntrack_info ctinfo;
 239         struct nf_conn *ct;
 240         u8 state = 0;
 241
 242         ct = nf_ct_get(skb, &ctinfo);
 243         if (ct) {
 244                 state = ovs_ct_get_state(ctinfo);
 245                 /* All unconfirmed entries are NEW connections. */
 246                 if (!nf_ct_is_confirmed(ct))
 247                         state |= OVS_CS_F_NEW;
 248                 /* OVS persists the related flag for the duration of the
 249                  * connection.
 250                  */
 251                 if (ct->master)
 252                         state |= OVS_CS_F_RELATED;
 253                 if (keep_nat_flags) {
 254                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 255                 } else {
 256                         if (ct->status & IPS_SRC_NAT)
 257                                 state |= OVS_CS_F_SRC_NAT;
 258                         if (ct->status & IPS_DST_NAT)
 259                                 state |= OVS_CS_F_DST_NAT;
 260                 }
 261                 zone = nf_ct_zone(ct);
 262         } else if (post_ct) {
 263                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 264                 if (info)
 265                         zone = &info->zone;
 266         }
 267         __ovs_ct_update_key(key, state, zone, ct);
 268 }
 269
 270 /* This is called to initialize CT key fields possibly coming in from the local
 271  * stack.
 272  */
 273 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 274 {
 275         ovs_ct_update_key(skb, NULL, key, false, false);
 276 }
 277
 278 #define IN6_ADDR_INITIALIZER(ADDR) \
 279         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 280           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 281
 282 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 283                    const struct sw_flow_key *output, struct sk_buff *skb)
 284 {
 285         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 286                 return -EMSGSIZE;
 287
 288         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 289             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 290                 return -EMSGSIZE;
 291
 292         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 293             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 294                 return -EMSGSIZE;
 295
 296         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 297             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 298                     &output->ct.labels))
 299                 return -EMSGSIZE;
 300
 301         if (swkey->ct_orig_proto) {
 302                 if (swkey->eth.type == htons(ETH_P_IP)) {
 303                         struct ovs_key_ct_tuple_ipv4 orig = {
 304                                 output->ipv4.ct_orig.src,
 305                                 output->ipv4.ct_orig.dst,
 306                                 output->ct.orig_tp.src,
 307                                 output->ct.orig_tp.dst,
 308                                 output->ct_orig_proto,
 309                         };
 310                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 311                                     sizeof(orig), &orig))
 312                                 return -EMSGSIZE;
 313                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 314                         struct ovs_key_ct_tuple_ipv6 orig = {
 315                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 316                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 317                                 output->ct.orig_tp.src,
 318                                 output->ct.orig_tp.dst,
 319                                 output->ct_orig_proto,
 320                         };
 321                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 322                                     sizeof(orig), &orig))
 323                                 return -EMSGSIZE;
 324                 }
 325         }
 326
 327         return 0;
 328 }
 329
 330 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 331                            u32 ct_mark, u32 mask)
 332 {
 333 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 334         u32 new_mark;
 335
 336         new_mark = ct_mark | (ct->mark & ~(mask));
 337         if (ct->mark != new_mark) {
 338                 ct->mark = new_mark;
 339                 if (nf_ct_is_confirmed(ct))
 340                         nf_conntrack_event_cache(IPCT_MARK, ct);
 341                 key->ct.mark = new_mark;
 342         }
 343
 344         return 0;
 345 #else
 346         return -ENOTSUPP;
 347 #endif
 348 }
 349
 350 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 351 {
 352         struct nf_conn_labels *cl;
 353
 354         cl = nf_ct_labels_find(ct);
 355         if (!cl) {
 356                 nf_ct_labels_ext_add(ct);
 357                 cl = nf_ct_labels_find(ct);
 358         }
 359
 360         return cl;
 361 }
 362
 363 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 364  * since the new connection is not yet confirmed, and thus no-one else has
 365  * access to it's labels, we simply write them over.
 366  */
 367 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 368                               const struct ovs_key_ct_labels *labels,
 369                               const struct ovs_key_ct_labels *mask)
 370 {
 371         struct nf_conn_labels *cl, *master_cl;
 372         bool have_mask = labels_nonzero(mask);
 373
 374         /* Inherit master's labels to the related connection? */
 375         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 376
 377         if (!master_cl && !have_mask)
 378                 return 0;   /* Nothing to do. */
 379
 380         cl = ovs_ct_get_conn_labels(ct);
 381         if (!cl)
 382                 return -ENOSPC;
 383
 384         /* Inherit the master's labels, if any. */
 385         if (master_cl)
 386                 *cl = *master_cl;
 387
 388         if (have_mask) {
 389                 u32 *dst = (u32 *)cl->bits;
 390                 int i;
 391
 392                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 393                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 394                                 (labels->ct_labels_32[i]
 395                                  & mask->ct_labels_32[i]);
 396         }
 397
 398         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 399          * IPCT_LABEL bit is set in the event cache.
 400          */
 401         nf_conntrack_event_cache(IPCT_LABEL, ct);
 402
 403         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 404
 405         return 0;
 406 }
 407
 408 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 409                              const struct ovs_key_ct_labels *labels,
 410                              const struct ovs_key_ct_labels *mask)
 411 {
 412         struct nf_conn_labels *cl;
 413         int err;
 414
 415         cl = ovs_ct_get_conn_labels(ct);
 416         if (!cl)
 417                 return -ENOSPC;
 418
 419         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 420                                     mask->ct_labels_32,
 421                                     OVS_CT_LABELS_LEN_32);
 422         if (err)
 423                 return err;
 424
 425         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 426
 427         return 0;
 428 }
 429
 430 /* 'skb' should already be pulled to nh_ofs. */
 431 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 432 {
 433         const struct nf_conntrack_helper *helper;
 434         const struct nf_conn_help *help;
 435         enum ip_conntrack_info ctinfo;
 436         unsigned int protoff;
 437         struct nf_conn *ct;
 438         int err;
 439
 440         ct = nf_ct_get(skb, &ctinfo);
 441         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 442                 return NF_ACCEPT;
 443
 444         help = nfct_help(ct);
 445         if (!help)
 446                 return NF_ACCEPT;
 447
 448         helper = rcu_dereference(help->helper);
 449         if (!helper)
 450                 return NF_ACCEPT;
 451
 452         switch (proto) {
 453         case NFPROTO_IPV4:
 454                 protoff = ip_hdrlen(skb);
 455                 break;
 456         case NFPROTO_IPV6: {
 457                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 458                 __be16 frag_off;
 459                 int ofs;
 460
 461                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 462                                        &frag_off);
 463                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 464                         pr_debug("proto header not found\n");
 465                         return NF_ACCEPT;
 466                 }
 467                 protoff = ofs;
 468                 break;
 469         }
 470         default:
 471                 WARN_ONCE(1, "helper invoked on non-IP family!");
 472                 return NF_DROP;
 473         }
 474
 475         err = helper->help(skb, protoff, ct, ctinfo);
 476         if (err != NF_ACCEPT)
 477                 return err;
 478
 479         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 480          * FTP with NAT) adusting the TCP payload size when mangling IP
 481          * addresses and/or port numbers in the text-based control connection.
 482          */
 483         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 484             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 485                 return NF_DROP;
 486         return NF_ACCEPT;
 487 }
 488
 489 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 490  * value if 'skb' is freed.
 491  */
 492 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 493                             u16 zone, struct sk_buff *skb)
 494 {
 495         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 496         int err;
 497
 498         if (key->eth.type == htons(ETH_P_IP)) {
 499                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 500
 501                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 502                 err = ip_defrag(net, skb, user);
 503                 if (err)
 504                         return err;
 505
 506                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 507 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 508         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 509                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 510
 511                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 512                 err = nf_ct_frag6_gather(net, skb, user);
 513                 if (err) {
 514                         if (err != -EINPROGRESS)
 515                                 kfree_skb(skb);
 516                         return err;
 517                 }
 518
 519                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 520                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 521 #endif
 522         } else {
 523                 kfree_skb(skb);
 524                 return -EPFNOSUPPORT;
 525         }
 526
 527         key->ip.frag = OVS_FRAG_TYPE_NONE;
 528         skb_clear_hash(skb);
 529         skb->ignore_df = 1;
 530         *OVS_CB(skb) = ovs_cb;
 531
 532         return 0;
 533 }
 534
 535 static struct nf_conntrack_expect *
 536 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 537                    u16 proto, const struct sk_buff *skb)
 538 {
 539         struct nf_conntrack_tuple tuple;
 540         struct nf_conntrack_expect *exp;
 541
 542         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 543                 return NULL;
 544
 545         exp = __nf_ct_expect_find(net, zone, &tuple);
 546         if (exp) {
 547                 struct nf_conntrack_tuple_hash *h;
 548
 549                 /* Delete existing conntrack entry, if it clashes with the
 550                  * expectation.  This can happen since conntrack ALGs do not
 551                  * check for clashes between (new) expectations and existing
 552                  * conntrack entries.  nf_conntrack_in() will check the
 553                  * expectations only if a conntrack entry can not be found,
 554                  * which can lead to OVS finding the expectation (here) in the
 555                  * init direction, but which will not be removed by the
 556                  * nf_conntrack_in() call, if a matching conntrack entry is
 557                  * found instead.  In this case all init direction packets
 558                  * would be reported as new related packets, while reply
 559                  * direction packets would be reported as un-related
 560                  * established packets.
 561                  */
 562                 h = nf_conntrack_find_get(net, zone, &tuple);
 563                 if (h) {
 564                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 565
 566                         nf_ct_delete(ct, 0, 0);
 567                         nf_conntrack_put(&ct->ct_general);
 568                 }
 569         }
 570
 571         return exp;
 572 }
 573
 574 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 575 static enum ip_conntrack_info
 576 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 577 {
 578         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 579
 580         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 581                 return IP_CT_ESTABLISHED_REPLY;
 582         /* Once we've had two way comms, always ESTABLISHED. */
 583         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 584                 return IP_CT_ESTABLISHED;
 585         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 586                 return IP_CT_RELATED;
 587         return IP_CT_NEW;
 588 }
 589
 590 /* Find an existing connection which this packet belongs to without
 591  * re-attributing statistics or modifying the connection state.  This allows an
 592  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 593  *
 594  * Must be called with rcu_read_lock.
 595  *
 596  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 597  * if there is no existing entry.
 598  */
 599 static struct nf_conn *
 600 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 601                      u8 l3num, struct sk_buff *skb, bool natted)
 602 {
 603         struct nf_conntrack_tuple tuple;
 604         struct nf_conntrack_tuple_hash *h;
 605         struct nf_conn *ct;
 606
 607         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 608                                net, &tuple)) {
 609                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 610                 return NULL;
 611         }
 612
 613         /* Must invert the tuple if skb has been transformed by NAT. */
 614         if (natted) {
 615                 struct nf_conntrack_tuple inverse;
 616
 617                 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
 618                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 619                         return NULL;
 620                 }
 621                 tuple = inverse;
 622         }
 623
 624         /* look for tuple match */
 625         h = nf_conntrack_find_get(net, zone, &tuple);
 626         if (!h)
 627                 return NULL;   /* Not found. */
 628
 629         ct = nf_ct_tuplehash_to_ctrack(h);
 630
 631         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 632          * select the other tuplehash to get the right 'ctinfo' bits for this
 633          * packet.
 634          */
 635         if (natted)
 636                 h = &ct->tuplehash[!h->tuple.dst.dir];
 637
 638         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 639         return ct;
 640 }
 641
 642 static
 643 struct nf_conn *ovs_ct_executed(struct net *net,
 644                                 const struct sw_flow_key *key,
 645                                 const struct ovs_conntrack_info *info,
 646                                 struct sk_buff *skb,
 647                                 bool *ct_executed)
 648 {
 649         struct nf_conn *ct = NULL;
 650
 651         /* If no ct, check if we have evidence that an existing conntrack entry
 652          * might be found for this skb.  This happens when we lose a skb->_nfct
 653          * due to an upcall, or if the direction is being forced.  If the
 654          * connection was not confirmed, it is not cached and needs to be run
 655          * through conntrack again.
 656          */
 657         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 658                        !(key->ct_state & OVS_CS_F_INVALID) &&
 659                        (key->ct_zone == info->zone.id);
 660
 661         if (*ct_executed || (!key->ct_state && info->force)) {
 662                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 663                                           !!(key->ct_state &
 664                                           OVS_CS_F_NAT_MASK));
 665         }
 666
 667         return ct;
 668 }
 669
 670 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 671 static bool skb_nfct_cached(struct net *net,
 672                             const struct sw_flow_key *key,
 673                             const struct ovs_conntrack_info *info,
 674                             struct sk_buff *skb)
 675 {
 676         enum ip_conntrack_info ctinfo;
 677         struct nf_conn *ct;
 678         bool ct_executed = true;
 679
 680         ct = nf_ct_get(skb, &ctinfo);
 681         if (!ct)
 682                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 683
 684         if (ct)
 685                 nf_ct_get(skb, &ctinfo);
 686         else
 687                 return false;
 688
 689         if (!net_eq(net, read_pnet(&ct->ct_net)))
 690                 return false;
 691         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 692                 return false;
 693         if (info->helper) {
 694                 struct nf_conn_help *help;
 695
 696                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 697                 if (help && rcu_access_pointer(help->helper) != info->helper)
 698                         return false;
 699         }
 700         /* Force conntrack entry direction to the current packet? */
 701         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 702                 /* Delete the conntrack entry if confirmed, else just release
 703                  * the reference.
 704                  */
 705                 if (nf_ct_is_confirmed(ct))
 706                         nf_ct_delete(ct, 0, 0);
 707
 708                 nf_conntrack_put(&ct->ct_general);
 709                 nf_ct_set(skb, NULL, 0);
 710                 return false;
 711         }
 712
 713         return ct_executed;
 714 }
 715
 716 #if IS_ENABLED(CONFIG_NF_NAT)
 717 /* Modelled after nf_nat_ipv[46]_fn().
 718  * range is only used for new, uninitialized NAT state.
 719  * Returns either NF_ACCEPT or NF_DROP.
 720  */
 721 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 722                               enum ip_conntrack_info ctinfo,
 723                               const struct nf_nat_range2 *range,
 724                               enum nf_nat_manip_type maniptype)
 725 {
 726         int hooknum, nh_off, err = NF_ACCEPT;
 727
 728         nh_off = skb_network_offset(skb);
 729         skb_pull_rcsum(skb, nh_off);
 730
 731         /* See HOOK2MANIP(). */
 732         if (maniptype == NF_NAT_MANIP_SRC)
 733                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 734         else
 735                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 736
 737         switch (ctinfo) {
 738         case IP_CT_RELATED:
 739         case IP_CT_RELATED_REPLY:
 740                 if (IS_ENABLED(CONFIG_NF_NAT) &&
 741                     skb->protocol == htons(ETH_P_IP) &&
 742                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 743                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 744                                                            hooknum))
 745                                 err = NF_DROP;
 746                         goto push;
 747                 } else if (IS_ENABLED(CONFIG_IPV6) &&
 748                            skb->protocol == htons(ETH_P_IPV6)) {
 749                         __be16 frag_off;
 750                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 751                         int hdrlen = ipv6_skip_exthdr(skb,
 752                                                       sizeof(struct ipv6hdr),
 753                                                       &nexthdr, &frag_off);
 754
 755                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 756                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 757                                                                      ctinfo,
 758                                                                      hooknum,
 759                                                                      hdrlen))
 760                                         err = NF_DROP;
 761                                 goto push;
 762                         }
 763                 }
 764                 /* Non-ICMP, fall thru to initialize if needed. */
 765                 /* fall through */
 766         case IP_CT_NEW:
 767                 /* Seen it before?  This can happen for loopback, retrans,
 768                  * or local packets.
 769                  */
 770                 if (!nf_nat_initialized(ct, maniptype)) {
 771                         /* Initialize according to the NAT action. */
 772                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 773                                 /* Action is set up to establish a new
 774                                  * mapping.
 775                                  */
 776                                 ? nf_nat_setup_info(ct, range, maniptype)
 777                                 : nf_nat_alloc_null_binding(ct, hooknum);
 778                         if (err != NF_ACCEPT)
 779                                 goto push;
 780                 }
 781                 break;
 782
 783         case IP_CT_ESTABLISHED:
 784         case IP_CT_ESTABLISHED_REPLY:
 785                 break;
 786
 787         default:
 788                 err = NF_DROP;
 789                 goto push;
 790         }
 791
 792         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 793 push:
 794         skb_push(skb, nh_off);
 795         skb_postpush_rcsum(skb, skb->data, nh_off);
 796
 797         return err;
 798 }
 799
 800 static void ovs_nat_update_key(struct sw_flow_key *key,
 801                                const struct sk_buff *skb,
 802                                enum nf_nat_manip_type maniptype)
 803 {
 804         if (maniptype == NF_NAT_MANIP_SRC) {
 805                 __be16 src;
 806
 807                 key->ct_state |= OVS_CS_F_SRC_NAT;
 808                 if (key->eth.type == htons(ETH_P_IP))
 809                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 810                 else if (key->eth.type == htons(ETH_P_IPV6))
 811                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 812                                sizeof(key->ipv6.addr.src));
 813                 else
 814                         return;
 815
 816                 if (key->ip.proto == IPPROTO_UDP)
 817                         src = udp_hdr(skb)->source;
 818                 else if (key->ip.proto == IPPROTO_TCP)
 819                         src = tcp_hdr(skb)->source;
 820                 else if (key->ip.proto == IPPROTO_SCTP)
 821                         src = sctp_hdr(skb)->source;
 822                 else
 823                         return;
 824
 825                 key->tp.src = src;
 826         } else {
 827                 __be16 dst;
 828
 829                 key->ct_state |= OVS_CS_F_DST_NAT;
 830                 if (key->eth.type == htons(ETH_P_IP))
 831                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 832                 else if (key->eth.type == htons(ETH_P_IPV6))
 833                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 834                                sizeof(key->ipv6.addr.dst));
 835                 else
 836                         return;
 837
 838                 if (key->ip.proto == IPPROTO_UDP)
 839                         dst = udp_hdr(skb)->dest;
 840                 else if (key->ip.proto == IPPROTO_TCP)
 841                         dst = tcp_hdr(skb)->dest;
 842                 else if (key->ip.proto == IPPROTO_SCTP)
 843                         dst = sctp_hdr(skb)->dest;
 844                 else
 845                         return;
 846
 847                 key->tp.dst = dst;
 848         }
 849 }
 850
 851 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 852 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 853                       const struct ovs_conntrack_info *info,
 854                       struct sk_buff *skb, struct nf_conn *ct,
 855                       enum ip_conntrack_info ctinfo)
 856 {
 857         enum nf_nat_manip_type maniptype;
 858         int err;
 859
 860         /* Add NAT extension if not confirmed yet. */
 861         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 862                 return NF_ACCEPT;   /* Can't NAT. */
 863
 864         /* Determine NAT type.
 865          * Check if the NAT type can be deduced from the tracked connection.
 866          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 867          * when committing.
 868          */
 869         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 870             ct->status & IPS_NAT_MASK &&
 871             (ctinfo != IP_CT_RELATED || info->commit)) {
 872                 /* NAT an established or related connection like before. */
 873                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 874                         /* This is the REPLY direction for a connection
 875                          * for which NAT was applied in the forward
 876                          * direction.  Do the reverse NAT.
 877                          */
 878                         maniptype = ct->status & IPS_SRC_NAT
 879                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 880                 else
 881                         maniptype = ct->status & IPS_SRC_NAT
 882                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 883         } else if (info->nat & OVS_CT_SRC_NAT) {
 884                 maniptype = NF_NAT_MANIP_SRC;
 885         } else if (info->nat & OVS_CT_DST_NAT) {
 886                 maniptype = NF_NAT_MANIP_DST;
 887         } else {
 888                 return NF_ACCEPT; /* Connection is not NATed. */
 889         }
 890         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 891
 892         /* Mark NAT done if successful and update the flow key. */
 893         if (err == NF_ACCEPT)
 894                 ovs_nat_update_key(key, skb, maniptype);
 895
 896         return err;
 897 }
 898 #else /* !CONFIG_NF_NAT */
 899 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 900                       const struct ovs_conntrack_info *info,
 901                       struct sk_buff *skb, struct nf_conn *ct,
 902                       enum ip_conntrack_info ctinfo)
 903 {
 904         return NF_ACCEPT;
 905 }
 906 #endif
 907
 908 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 909  * not done already.  Update key with new CT state after passing the packet
 910  * through conntrack.
 911  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 912  * set to NULL and 0 will be returned.
 913  */
 914 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 915                            const struct ovs_conntrack_info *info,
 916                            struct sk_buff *skb)
 917 {
 918         /* If we are recirculating packets to match on conntrack fields and
 919          * committing with a separate conntrack action,  then we don't need to
 920          * actually run the packet through conntrack twice unless it's for a
 921          * different zone.
 922          */
 923         bool cached = skb_nfct_cached(net, key, info, skb);
 924         enum ip_conntrack_info ctinfo;
 925         struct nf_conn *ct;
 926
 927         if (!cached) {
 928                 struct nf_hook_state state = {
 929                         .hook = NF_INET_PRE_ROUTING,
 930                         .pf = info->family,
 931                         .net = net,
 932                 };
 933                 struct nf_conn *tmpl = info->ct;
 934                 int err;
 935
 936                 /* Associate skb with specified zone. */
 937                 if (tmpl) {
 938                         if (skb_nfct(skb))
 939                                 nf_conntrack_put(skb_nfct(skb));
 940                         nf_conntrack_get(&tmpl->ct_general);
 941                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 942                 }
 943
 944                 err = nf_conntrack_in(skb, &state);
 945                 if (err != NF_ACCEPT)
 946                         return -ENOENT;
 947
 948                 /* Clear CT state NAT flags to mark that we have not yet done
 949                  * NAT after the nf_conntrack_in() call.  We can actually clear
 950                  * the whole state, as it will be re-initialized below.
 951                  */
 952                 key->ct_state = 0;
 953
 954                 /* Update the key, but keep the NAT flags. */
 955                 ovs_ct_update_key(skb, info, key, true, true);
 956         }
 957
 958         ct = nf_ct_get(skb, &ctinfo);
 959         if (ct) {
 960                 /* Packets starting a new connection must be NATted before the
 961                  * helper, so that the helper knows about the NAT.  We enforce
 962                  * this by delaying both NAT and helper calls for unconfirmed
 963                  * connections until the committing CT action.  For later
 964                  * packets NAT and Helper may be called in either order.
 965                  *
 966                  * NAT will be done only if the CT action has NAT, and only
 967                  * once per packet (per zone), as guarded by the NAT bits in
 968                  * the key->ct_state.
 969                  */
 970                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 971                     (nf_ct_is_confirmed(ct) || info->commit) &&
 972                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 973                         return -EINVAL;
 974                 }
 975
 976                 /* Userspace may decide to perform a ct lookup without a helper
 977                  * specified followed by a (recirculate and) commit with one.
 978                  * Therefore, for unconfirmed connections which we will commit,
 979                  * we need to attach the helper here.
 980                  */
 981                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 982                     info->helper && !nfct_help(ct)) {
 983                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 984                                                             GFP_ATOMIC);
 985                         if (err)
 986                                 return err;
 987
 988                         /* helper installed, add seqadj if NAT is required */
 989                         if (info->nat && !nfct_seqadj(ct)) {
 990                                 if (!nfct_seqadj_ext_add(ct))
 991                                         return -EINVAL;
 992                         }
 993                 }
 994
 995                 /* Call the helper only if:
 996                  * - nf_conntrack_in() was executed above ("!cached") for a
 997                  *   confirmed connection, or
 998                  * - When committing an unconfirmed connection.
 999                  */
1000                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1001                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1002                         return -EINVAL;
1003                 }
1004         }
1005
1006         return 0;
1007 }
1008
1009 /* Lookup connection and read fields into key. */
1010 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1011                          const struct ovs_conntrack_info *info,
1012                          struct sk_buff *skb)
1013 {
1014         struct nf_conntrack_expect *exp;
1015
1016         /* If we pass an expected packet through nf_conntrack_in() the
1017          * expectation is typically removed, but the packet could still be
1018          * lost in upcall processing.  To prevent this from happening we
1019          * perform an explicit expectation lookup.  Expected connections are
1020          * always new, and will be passed through conntrack only when they are
1021          * committed, as it is OK to remove the expectation at that time.
1022          */
1023         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1024         if (exp) {
1025                 u8 state;
1026
1027                 /* NOTE: New connections are NATted and Helped only when
1028                  * committed, so we are not calling into NAT here.
1029                  */
1030                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1031                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1032         } else {
1033                 struct nf_conn *ct;
1034                 int err;
1035
1036                 err = __ovs_ct_lookup(net, key, info, skb);
1037                 if (err)
1038                         return err;
1039
1040                 ct = (struct nf_conn *)skb_nfct(skb);
1041                 if (ct)
1042                         nf_ct_deliver_cached_events(ct);
1043         }
1044
1045         return 0;
1046 }
1047
1048 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1049 {
1050         size_t i;
1051
1052         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1053                 if (labels->ct_labels_32[i])
1054                         return true;
1055
1056         return false;
1057 }
1058
1059 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1060 static struct hlist_head *ct_limit_hash_bucket(
1061         const struct ovs_ct_limit_info *info, u16 zone)
1062 {
1063         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1064 }
1065
1066 /* Call with ovs_mutex */
1067 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1068                          struct ovs_ct_limit *new_ct_limit)
1069 {
1070         struct ovs_ct_limit *ct_limit;
1071         struct hlist_head *head;
1072
1073         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1074         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1075                 if (ct_limit->zone == new_ct_limit->zone) {
1076                         hlist_replace_rcu(&ct_limit->hlist_node,
1077                                           &new_ct_limit->hlist_node);
1078                         kfree_rcu(ct_limit, rcu);
1079                         return;
1080                 }
1081         }
1082
1083         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1084 }
1085
1086 /* Call with ovs_mutex */
1087 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1088 {
1089         struct ovs_ct_limit *ct_limit;
1090         struct hlist_head *head;
1091         struct hlist_node *n;
1092
1093         head = ct_limit_hash_bucket(info, zone);
1094         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1095                 if (ct_limit->zone == zone) {
1096                         hlist_del_rcu(&ct_limit->hlist_node);
1097                         kfree_rcu(ct_limit, rcu);
1098                         return;
1099                 }
1100         }
1101 }
1102
1103 /* Call with RCU read lock */
1104 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1105 {
1106         struct ovs_ct_limit *ct_limit;
1107         struct hlist_head *head;
1108
1109         head = ct_limit_hash_bucket(info, zone);
1110         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1111                 if (ct_limit->zone == zone)
1112                         return ct_limit->limit;
1113         }
1114
1115         return info->default_limit;
1116 }
1117
1118 static int ovs_ct_check_limit(struct net *net,
1119                               const struct ovs_conntrack_info *info,
1120                               const struct nf_conntrack_tuple *tuple)
1121 {
1122         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1123         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1124         u32 per_zone_limit, connections;
1125         u32 conncount_key;
1126
1127         conncount_key = info->zone.id;
1128
1129         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1130         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1131                 return 0;
1132
1133         connections = nf_conncount_count(net, ct_limit_info->data,
1134                                          &conncount_key, tuple, &info->zone);
1135         if (connections > per_zone_limit)
1136                 return -ENOMEM;
1137
1138         return 0;
1139 }
1140 #endif
1141
1142 /* Lookup connection and confirm if unconfirmed. */
1143 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1144                          const struct ovs_conntrack_info *info,
1145                          struct sk_buff *skb)
1146 {
1147         enum ip_conntrack_info ctinfo;
1148         struct nf_conn *ct;
1149         int err;
1150
1151         err = __ovs_ct_lookup(net, key, info, skb);
1152         if (err)
1153                 return err;
1154
1155         /* The connection could be invalid, in which case this is a no-op.*/
1156         ct = nf_ct_get(skb, &ctinfo);
1157         if (!ct)
1158                 return 0;
1159
1160 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1161         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1162                 if (!nf_ct_is_confirmed(ct)) {
1163                         err = ovs_ct_check_limit(net, info,
1164                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1165                         if (err) {
1166                                 net_warn_ratelimited("openvswitch: zone: %u "
1167                                         "exceeds conntrack limit\n",
1168                                         info->zone.id);
1169                                 return err;
1170                         }
1171                 }
1172         }
1173 #endif
1174
1175         /* Set the conntrack event mask if given.  NEW and DELETE events have
1176          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1177          * typically would receive many kinds of updates.  Setting the event
1178          * mask allows those events to be filtered.  The set event mask will
1179          * remain in effect for the lifetime of the connection unless changed
1180          * by a further CT action with both the commit flag and the eventmask
1181          * option. */
1182         if (info->have_eventmask) {
1183                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1184
1185                 if (cache)
1186                         cache->ctmask = info->eventmask;
1187         }
1188
1189         /* Apply changes before confirming the connection so that the initial
1190          * conntrack NEW netlink event carries the values given in the CT
1191          * action.
1192          */
1193         if (info->mark.mask) {
1194                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1195                                       info->mark.mask);
1196                 if (err)
1197                         return err;
1198         }
1199         if (!nf_ct_is_confirmed(ct)) {
1200                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1201                                          &info->labels.mask);
1202                 if (err)
1203                         return err;
1204         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1205                    labels_nonzero(&info->labels.mask)) {
1206                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1207                                         &info->labels.mask);
1208                 if (err)
1209                         return err;
1210         }
1211         /* This will take care of sending queued events even if the connection
1212          * is already confirmed.
1213          */
1214         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1215                 return -EINVAL;
1216
1217         return 0;
1218 }
1219
1220 /* Trim the skb to the length specified by the IP/IPv6 header,
1221  * removing any trailing lower-layer padding. This prepares the skb
1222  * for higher-layer processing that assumes skb->len excludes padding
1223  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1224  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1225  */
1226 static int ovs_skb_network_trim(struct sk_buff *skb)
1227 {
1228         unsigned int len;
1229         int err;
1230
1231         switch (skb->protocol) {
1232         case htons(ETH_P_IP):
1233                 len = ntohs(ip_hdr(skb)->tot_len);
1234                 break;
1235         case htons(ETH_P_IPV6):
1236                 len = sizeof(struct ipv6hdr)
1237                         + ntohs(ipv6_hdr(skb)->payload_len);
1238                 break;
1239         default:
1240                 len = skb->len;
1241         }
1242
1243         err = pskb_trim_rcsum(skb, len);
1244         if (err)
1245                 kfree_skb(skb);
1246
1247         return err;
1248 }
1249
1250 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1251  * value if 'skb' is freed.
1252  */
1253 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1254                    struct sw_flow_key *key,
1255                    const struct ovs_conntrack_info *info)
1256 {
1257         int nh_ofs;
1258         int err;
1259
1260         /* The conntrack module expects to be working at L3. */
1261         nh_ofs = skb_network_offset(skb);
1262         skb_pull_rcsum(skb, nh_ofs);
1263
1264         err = ovs_skb_network_trim(skb);
1265         if (err)
1266                 return err;
1267
1268         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1269                 err = handle_fragments(net, key, info->zone.id, skb);
1270                 if (err)
1271                         return err;
1272         }
1273
1274         if (info->commit)
1275                 err = ovs_ct_commit(net, key, info, skb);
1276         else
1277                 err = ovs_ct_lookup(net, key, info, skb);
1278
1279         skb_push(skb, nh_ofs);
1280         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1281         if (err)
1282                 kfree_skb(skb);
1283         return err;
1284 }
1285
1286 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1287 {
1288         if (skb_nfct(skb)) {
1289                 nf_conntrack_put(skb_nfct(skb));
1290                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1291                 ovs_ct_fill_key(skb, key);
1292         }
1293
1294         return 0;
1295 }
1296
1297 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1298                              const struct sw_flow_key *key, bool log)
1299 {
1300         struct nf_conntrack_helper *helper;
1301         struct nf_conn_help *help;
1302         int ret = 0;
1303
1304         helper = nf_conntrack_helper_try_module_get(name, info->family,
1305                                                     key->ip.proto);
1306         if (!helper) {
1307                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1308                 return -EINVAL;
1309         }
1310
1311         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1312         if (!help) {
1313                 nf_conntrack_helper_put(helper);
1314                 return -ENOMEM;
1315         }
1316
1317 #if IS_ENABLED(CONFIG_NF_NAT)
1318         if (info->nat) {
1319                 ret = nf_nat_helper_try_module_get(name, info->family,
1320                                                    key->ip.proto);
1321                 if (ret) {
1322                         nf_conntrack_helper_put(helper);
1323                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1324                                   name, ret);
1325                         return ret;
1326                 }
1327         }
1328 #endif
1329         rcu_assign_pointer(help->helper, helper);
1330         info->helper = helper;
1331         return ret;
1332 }
1333
1334 #if IS_ENABLED(CONFIG_NF_NAT)
1335 static int parse_nat(const struct nlattr *attr,
1336                      struct ovs_conntrack_info *info, bool log)
1337 {
1338         struct nlattr *a;
1339         int rem;
1340         bool have_ip_max = false;
1341         bool have_proto_max = false;
1342         bool ip_vers = (info->family == NFPROTO_IPV6);
1343
1344         nla_for_each_nested(a, attr, rem) {
1345                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1346                         [OVS_NAT_ATTR_SRC] = {0, 0},
1347                         [OVS_NAT_ATTR_DST] = {0, 0},
1348                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1349                                                  sizeof(struct in6_addr)},
1350                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1351                                                  sizeof(struct in6_addr)},
1352                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1353                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1354                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1355                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1356                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1357                 };
1358                 int type = nla_type(a);
1359
1360                 if (type > OVS_NAT_ATTR_MAX) {
1361                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1362                                   type, OVS_NAT_ATTR_MAX);
1363                         return -EINVAL;
1364                 }
1365
1366                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1367                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1368                                   type, nla_len(a),
1369                                   ovs_nat_attr_lens[type][ip_vers]);
1370                         return -EINVAL;
1371                 }
1372
1373                 switch (type) {
1374                 case OVS_NAT_ATTR_SRC:
1375                 case OVS_NAT_ATTR_DST:
1376                         if (info->nat) {
1377                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1378                                 return -ERANGE;
1379                         }
1380                         info->nat |= OVS_CT_NAT;
1381                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1382                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1383                         break;
1384
1385                 case OVS_NAT_ATTR_IP_MIN:
1386                         nla_memcpy(&info->range.min_addr, a,
1387                                    sizeof(info->range.min_addr));
1388                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1389                         break;
1390
1391                 case OVS_NAT_ATTR_IP_MAX:
1392                         have_ip_max = true;
1393                         nla_memcpy(&info->range.max_addr, a,
1394                                    sizeof(info->range.max_addr));
1395                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1396                         break;
1397
1398                 case OVS_NAT_ATTR_PROTO_MIN:
1399                         info->range.min_proto.all = htons(nla_get_u16(a));
1400                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1401                         break;
1402
1403                 case OVS_NAT_ATTR_PROTO_MAX:
1404                         have_proto_max = true;
1405                         info->range.max_proto.all = htons(nla_get_u16(a));
1406                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1407                         break;
1408
1409                 case OVS_NAT_ATTR_PERSISTENT:
1410                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1411                         break;
1412
1413                 case OVS_NAT_ATTR_PROTO_HASH:
1414                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1415                         break;
1416
1417                 case OVS_NAT_ATTR_PROTO_RANDOM:
1418                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1419                         break;
1420
1421                 default:
1422                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1423                         return -EINVAL;
1424                 }
1425         }
1426
1427         if (rem > 0) {
1428                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1429                 return -EINVAL;
1430         }
1431         if (!info->nat) {
1432                 /* Do not allow flags if no type is given. */
1433                 if (info->range.flags) {
1434                         OVS_NLERR(log,
1435                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1436                                   );
1437                         return -EINVAL;
1438                 }
1439                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1440         } else if (!info->commit) {
1441                 OVS_NLERR(log,
1442                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1443                           );
1444                 return -EINVAL;
1445         }
1446         /* Allow missing IP_MAX. */
1447         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1448                 memcpy(&info->range.max_addr, &info->range.min_addr,
1449                        sizeof(info->range.max_addr));
1450         }
1451         /* Allow missing PROTO_MAX. */
1452         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1453             !have_proto_max) {
1454                 info->range.max_proto.all = info->range.min_proto.all;
1455         }
1456         return 0;
1457 }
1458 #endif
1459
1460 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1461         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1462         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1463         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1464                                     .maxlen = sizeof(u16) },
1465         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1466                                     .maxlen = sizeof(struct md_mark) },
1467         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1468                                     .maxlen = sizeof(struct md_labels) },
1469         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1470                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1471 #if IS_ENABLED(CONFIG_NF_NAT)
1472         /* NAT length is checked when parsing the nested attributes. */
1473         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1474 #endif
1475         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1476                                     .maxlen = sizeof(u32) },
1477         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1478                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1479 };
1480
1481 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1482                     const char **helper, bool log)
1483 {
1484         struct nlattr *a;
1485         int rem;
1486
1487         nla_for_each_nested(a, attr, rem) {
1488                 int type = nla_type(a);
1489                 int maxlen;
1490                 int minlen;
1491
1492                 if (type > OVS_CT_ATTR_MAX) {
1493                         OVS_NLERR(log,
1494                                   "Unknown conntrack attr (type=%d, max=%d)",
1495                                   type, OVS_CT_ATTR_MAX);
1496                         return -EINVAL;
1497                 }
1498
1499                 maxlen = ovs_ct_attr_lens[type].maxlen;
1500                 minlen = ovs_ct_attr_lens[type].minlen;
1501                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1502                         OVS_NLERR(log,
1503                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1504                                   type, nla_len(a), maxlen);
1505                         return -EINVAL;
1506                 }
1507
1508                 switch (type) {
1509                 case OVS_CT_ATTR_FORCE_COMMIT:
1510                         info->force = true;
1511                         /* fall through. */
1512                 case OVS_CT_ATTR_COMMIT:
1513                         info->commit = true;
1514                         break;
1515 #ifdef CONFIG_NF_CONNTRACK_ZONES
1516                 case OVS_CT_ATTR_ZONE:
1517                         info->zone.id = nla_get_u16(a);
1518                         break;
1519 #endif
1520 #ifdef CONFIG_NF_CONNTRACK_MARK
1521                 case OVS_CT_ATTR_MARK: {
1522                         struct md_mark *mark = nla_data(a);
1523
1524                         if (!mark->mask) {
1525                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1526                                 return -EINVAL;
1527                         }
1528                         info->mark = *mark;
1529                         break;
1530                 }
1531 #endif
1532 #ifdef CONFIG_NF_CONNTRACK_LABELS
1533                 case OVS_CT_ATTR_LABELS: {
1534                         struct md_labels *labels = nla_data(a);
1535
1536                         if (!labels_nonzero(&labels->mask)) {
1537                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1538                                 return -EINVAL;
1539                         }
1540                         info->labels = *labels;
1541                         break;
1542                 }
1543 #endif
1544                 case OVS_CT_ATTR_HELPER:
1545                         *helper = nla_data(a);
1546                         if (!memchr(*helper, '\0', nla_len(a))) {
1547                                 OVS_NLERR(log, "Invalid conntrack helper");
1548                                 return -EINVAL;
1549                         }
1550                         break;
1551 #if IS_ENABLED(CONFIG_NF_NAT)
1552                 case OVS_CT_ATTR_NAT: {
1553                         int err = parse_nat(a, info, log);
1554
1555                         if (err)
1556                                 return err;
1557                         break;
1558                 }
1559 #endif
1560                 case OVS_CT_ATTR_EVENTMASK:
1561                         info->have_eventmask = true;
1562                         info->eventmask = nla_get_u32(a);
1563                         break;
1564 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1565                 case OVS_CT_ATTR_TIMEOUT:
1566                         memcpy(info->timeout, nla_data(a), nla_len(a));
1567                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1568                                 OVS_NLERR(log, "Invalid conntrack helper");
1569                                 return -EINVAL;
1570                         }
1571                         break;
1572 #endif
1573
1574                 default:
1575                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1576                                   type);
1577                         return -EINVAL;
1578                 }
1579         }
1580
1581 #ifdef CONFIG_NF_CONNTRACK_MARK
1582         if (!info->commit && info->mark.mask) {
1583                 OVS_NLERR(log,
1584                           "Setting conntrack mark requires 'commit' flag.");
1585                 return -EINVAL;
1586         }
1587 #endif
1588 #ifdef CONFIG_NF_CONNTRACK_LABELS
1589         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1590                 OVS_NLERR(log,
1591                           "Setting conntrack labels requires 'commit' flag.");
1592                 return -EINVAL;
1593         }
1594 #endif
1595         if (rem > 0) {
1596                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1597                 return -EINVAL;
1598         }
1599
1600         return 0;
1601 }
1602
1603 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1604 {
1605         if (attr == OVS_KEY_ATTR_CT_STATE)
1606                 return true;
1607         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1608             attr == OVS_KEY_ATTR_CT_ZONE)
1609                 return true;
1610         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1611             attr == OVS_KEY_ATTR_CT_MARK)
1612                 return true;
1613         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1614             attr == OVS_KEY_ATTR_CT_LABELS) {
1615                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1616
1617                 return ovs_net->xt_label;
1618         }
1619
1620         return false;
1621 }
1622
1623 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1624                        const struct sw_flow_key *key,
1625                        struct sw_flow_actions **sfa,  bool log)
1626 {
1627         struct ovs_conntrack_info ct_info;
1628         const char *helper = NULL;
1629         u16 family;
1630         int err;
1631
1632         family = key_to_nfproto(key);
1633         if (family == NFPROTO_UNSPEC) {
1634                 OVS_NLERR(log, "ct family unspecified");
1635                 return -EINVAL;
1636         }
1637
1638         memset(&ct_info, 0, sizeof(ct_info));
1639         ct_info.family = family;
1640
1641         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1642                         NF_CT_DEFAULT_ZONE_DIR, 0);
1643
1644         err = parse_ct(attr, &ct_info, &helper, log);
1645         if (err)
1646                 return err;
1647
1648         /* Set up template for tracking connections in specific zones. */
1649         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1650         if (!ct_info.ct) {
1651                 OVS_NLERR(log, "Failed to allocate conntrack template");
1652                 return -ENOMEM;
1653         }
1654
1655         if (ct_info.timeout[0]) {
1656                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1657                                       ct_info.timeout))
1658                         pr_info_ratelimited("Failed to associated timeout "
1659                                             "policy `%s'\n", ct_info.timeout);
1660         }
1661
1662         if (helper) {
1663                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1664                 if (err)
1665                         goto err_free_ct;
1666         }
1667
1668         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1669                                  sizeof(ct_info), log);
1670         if (err)
1671                 goto err_free_ct;
1672
1673         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1674         nf_conntrack_get(&ct_info.ct->ct_general);
1675         return 0;
1676 err_free_ct:
1677         __ovs_ct_free_action(&ct_info);
1678         return err;
1679 }
1680
1681 #if IS_ENABLED(CONFIG_NF_NAT)
1682 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1683                                struct sk_buff *skb)
1684 {
1685         struct nlattr *start;
1686
1687         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1688         if (!start)
1689                 return false;
1690
1691         if (info->nat & OVS_CT_SRC_NAT) {
1692                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1693                         return false;
1694         } else if (info->nat & OVS_CT_DST_NAT) {
1695                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1696                         return false;
1697         } else {
1698                 goto out;
1699         }
1700
1701         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1702                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1703                     info->family == NFPROTO_IPV4) {
1704                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1705                                             info->range.min_addr.ip) ||
1706                             (info->range.max_addr.ip
1707                              != info->range.min_addr.ip &&
1708                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1709                                               info->range.max_addr.ip))))
1710                                 return false;
1711                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1712                            info->family == NFPROTO_IPV6) {
1713                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1714                                              &info->range.min_addr.in6) ||
1715                             (memcmp(&info->range.max_addr.in6,
1716                                     &info->range.min_addr.in6,
1717                                     sizeof(info->range.max_addr.in6)) &&
1718                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1719                                                &info->range.max_addr.in6))))
1720                                 return false;
1721                 } else {
1722                         return false;
1723                 }
1724         }
1725         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1726             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1727                          ntohs(info->range.min_proto.all)) ||
1728              (info->range.max_proto.all != info->range.min_proto.all &&
1729               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1730                           ntohs(info->range.max_proto.all)))))
1731                 return false;
1732
1733         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1734             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1735                 return false;
1736         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1737             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1738                 return false;
1739         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1740             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1741                 return false;
1742 out:
1743         nla_nest_end(skb, start);
1744
1745         return true;
1746 }
1747 #endif
1748
1749 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1750                           struct sk_buff *skb)
1751 {
1752         struct nlattr *start;
1753
1754         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1755         if (!start)
1756                 return -EMSGSIZE;
1757
1758         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1759                                             ? OVS_CT_ATTR_FORCE_COMMIT
1760                                             : OVS_CT_ATTR_COMMIT))
1761                 return -EMSGSIZE;
1762         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1763             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1764                 return -EMSGSIZE;
1765         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1766             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1767                     &ct_info->mark))
1768                 return -EMSGSIZE;
1769         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1770             labels_nonzero(&ct_info->labels.mask) &&
1771             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1772                     &ct_info->labels))
1773                 return -EMSGSIZE;
1774         if (ct_info->helper) {
1775                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1776                                    ct_info->helper->name))
1777                         return -EMSGSIZE;
1778         }
1779         if (ct_info->have_eventmask &&
1780             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1781                 return -EMSGSIZE;
1782         if (ct_info->timeout[0]) {
1783                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1784                         return -EMSGSIZE;
1785         }
1786
1787 #if IS_ENABLED(CONFIG_NF_NAT)
1788         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1789                 return -EMSGSIZE;
1790 #endif
1791         nla_nest_end(skb, start);
1792
1793         return 0;
1794 }
1795
1796 void ovs_ct_free_action(const struct nlattr *a)
1797 {
1798         struct ovs_conntrack_info *ct_info = nla_data(a);
1799
1800         __ovs_ct_free_action(ct_info);
1801 }
1802
1803 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1804 {
1805         if (ct_info->helper) {
1806 #if IS_ENABLED(CONFIG_NF_NAT)
1807                 if (ct_info->nat)
1808                         nf_nat_helper_put(ct_info->helper);
1809 #endif
1810                 nf_conntrack_helper_put(ct_info->helper);
1811         }
1812         if (ct_info->ct) {
1813                 if (ct_info->timeout[0])
1814                         nf_ct_destroy_timeout(ct_info->ct);
1815                 nf_ct_tmpl_free(ct_info->ct);
1816         }
1817 }
1818
1819 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1820 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1821 {
1822         int i, err;
1823
1824         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1825                                          GFP_KERNEL);
1826         if (!ovs_net->ct_limit_info)
1827                 return -ENOMEM;
1828
1829         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1830         ovs_net->ct_limit_info->limits =
1831                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1832                               GFP_KERNEL);
1833         if (!ovs_net->ct_limit_info->limits) {
1834                 kfree(ovs_net->ct_limit_info);
1835                 return -ENOMEM;
1836         }
1837
1838         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1839                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1840
1841         ovs_net->ct_limit_info->data =
1842                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1843
1844         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1845                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1846                 kfree(ovs_net->ct_limit_info->limits);
1847                 kfree(ovs_net->ct_limit_info);
1848                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1849                 return err;
1850         }
1851         return 0;
1852 }
1853
1854 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1855 {
1856         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1857         int i;
1858
1859         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1860         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1861                 struct hlist_head *head = &info->limits[i];
1862                 struct ovs_ct_limit *ct_limit;
1863
1864                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1865                         kfree_rcu(ct_limit, rcu);
1866         }
1867         kfree(ovs_net->ct_limit_info->limits);
1868         kfree(ovs_net->ct_limit_info);
1869 }
1870
1871 static struct sk_buff *
1872 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1873                              struct ovs_header **ovs_reply_header)
1874 {
1875         struct ovs_header *ovs_header = info->userhdr;
1876         struct sk_buff *skb;
1877
1878         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1879         if (!skb)
1880                 return ERR_PTR(-ENOMEM);
1881
1882         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1883                                         info->snd_seq,
1884                                         &dp_ct_limit_genl_family, 0, cmd);
1885
1886         if (!*ovs_reply_header) {
1887                 nlmsg_free(skb);
1888                 return ERR_PTR(-EMSGSIZE);
1889         }
1890         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1891
1892         return skb;
1893 }
1894
1895 static bool check_zone_id(int zone_id, u16 *pzone)
1896 {
1897         if (zone_id >= 0 && zone_id <= 65535) {
1898                 *pzone = (u16)zone_id;
1899                 return true;
1900         }
1901         return false;
1902 }
1903
1904 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1905                                        struct ovs_ct_limit_info *info)
1906 {
1907         struct ovs_zone_limit *zone_limit;
1908         int rem;
1909         u16 zone;
1910
1911         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1912         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1913
1914         while (rem >= sizeof(*zone_limit)) {
1915                 if (unlikely(zone_limit->zone_id ==
1916                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1917                         ovs_lock();
1918                         info->default_limit = zone_limit->limit;
1919                         ovs_unlock();
1920                 } else if (unlikely(!check_zone_id(
1921                                 zone_limit->zone_id, &zone))) {
1922                         OVS_NLERR(true, "zone id is out of range");
1923                 } else {
1924                         struct ovs_ct_limit *ct_limit;
1925
1926                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1927                         if (!ct_limit)
1928                                 return -ENOMEM;
1929
1930                         ct_limit->zone = zone;
1931                         ct_limit->limit = zone_limit->limit;
1932
1933                         ovs_lock();
1934                         ct_limit_set(info, ct_limit);
1935                         ovs_unlock();
1936                 }
1937                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1938                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1939                                 NLA_ALIGN(sizeof(*zone_limit)));
1940         }
1941
1942         if (rem)
1943                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1944
1945         return 0;
1946 }
1947
1948 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1949                                        struct ovs_ct_limit_info *info)
1950 {
1951         struct ovs_zone_limit *zone_limit;
1952         int rem;
1953         u16 zone;
1954
1955         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1956         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1957
1958         while (rem >= sizeof(*zone_limit)) {
1959                 if (unlikely(zone_limit->zone_id ==
1960                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1961                         ovs_lock();
1962                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1963                         ovs_unlock();
1964                 } else if (unlikely(!check_zone_id(
1965                                 zone_limit->zone_id, &zone))) {
1966                         OVS_NLERR(true, "zone id is out of range");
1967                 } else {
1968                         ovs_lock();
1969                         ct_limit_del(info, zone);
1970                         ovs_unlock();
1971                 }
1972                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1973                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1974                                 NLA_ALIGN(sizeof(*zone_limit)));
1975         }
1976
1977         if (rem)
1978                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1979
1980         return 0;
1981 }
1982
1983 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1984                                           struct sk_buff *reply)
1985 {
1986         struct ovs_zone_limit zone_limit;
1987         int err;
1988
1989         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1990         zone_limit.limit = info->default_limit;
1991         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1992         if (err)
1993                 return err;
1994
1995         return 0;
1996 }
1997
1998 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1999                                          struct nf_conncount_data *data,
2000                                          u16 zone_id, u32 limit,
2001                                          struct sk_buff *reply)
2002 {
2003         struct nf_conntrack_zone ct_zone;
2004         struct ovs_zone_limit zone_limit;
2005         u32 conncount_key = zone_id;
2006
2007         zone_limit.zone_id = zone_id;
2008         zone_limit.limit = limit;
2009         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2010
2011         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2012                                               &ct_zone);
2013         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2014 }
2015
2016 static int ovs_ct_limit_get_zone_limit(struct net *net,
2017                                        struct nlattr *nla_zone_limit,
2018                                        struct ovs_ct_limit_info *info,
2019                                        struct sk_buff *reply)
2020 {
2021         struct ovs_zone_limit *zone_limit;
2022         int rem, err;
2023         u32 limit;
2024         u16 zone;
2025
2026         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2027         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2028
2029         while (rem >= sizeof(*zone_limit)) {
2030                 if (unlikely(zone_limit->zone_id ==
2031                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2032                         err = ovs_ct_limit_get_default_limit(info, reply);
2033                         if (err)
2034                                 return err;
2035                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2036                                                         &zone))) {
2037                         OVS_NLERR(true, "zone id is out of range");
2038                 } else {
2039                         rcu_read_lock();
2040                         limit = ct_limit_get(info, zone);
2041                         rcu_read_unlock();
2042
2043                         err = __ovs_ct_limit_get_zone_limit(
2044                                 net, info->data, zone, limit, reply);
2045                         if (err)
2046                                 return err;
2047                 }
2048                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2049                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2050                                 NLA_ALIGN(sizeof(*zone_limit)));
2051         }
2052
2053         if (rem)
2054                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2055
2056         return 0;
2057 }
2058
2059 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2060                                            struct ovs_ct_limit_info *info,
2061                                            struct sk_buff *reply)
2062 {
2063         struct ovs_ct_limit *ct_limit;
2064         struct hlist_head *head;
2065         int i, err = 0;
2066
2067         err = ovs_ct_limit_get_default_limit(info, reply);
2068         if (err)
2069                 return err;
2070
2071         rcu_read_lock();
2072         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2073                 head = &info->limits[i];
2074                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2075                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2076                                 ct_limit->zone, ct_limit->limit, reply);
2077                         if (err)
2078                                 goto exit_err;
2079                 }
2080         }
2081
2082 exit_err:
2083         rcu_read_unlock();
2084         return err;
2085 }
2086
2087 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2088 {
2089         struct nlattr **a = info->attrs;
2090         struct sk_buff *reply;
2091         struct ovs_header *ovs_reply_header;
2092         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2093         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2094         int err;
2095
2096         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2097                                              &ovs_reply_header);
2098         if (IS_ERR(reply))
2099                 return PTR_ERR(reply);
2100
2101         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2102                 err = -EINVAL;
2103                 goto exit_err;
2104         }
2105
2106         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2107                                           ct_limit_info);
2108         if (err)
2109                 goto exit_err;
2110
2111         static_branch_enable(&ovs_ct_limit_enabled);
2112
2113         genlmsg_end(reply, ovs_reply_header);
2114         return genlmsg_reply(reply, info);
2115
2116 exit_err:
2117         nlmsg_free(reply);
2118         return err;
2119 }
2120
2121 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2122 {
2123         struct nlattr **a = info->attrs;
2124         struct sk_buff *reply;
2125         struct ovs_header *ovs_reply_header;
2126         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2127         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2128         int err;
2129
2130         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2131                                              &ovs_reply_header);
2132         if (IS_ERR(reply))
2133                 return PTR_ERR(reply);
2134
2135         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2136                 err = -EINVAL;
2137                 goto exit_err;
2138         }
2139
2140         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2141                                           ct_limit_info);
2142         if (err)
2143                 goto exit_err;
2144
2145         genlmsg_end(reply, ovs_reply_header);
2146         return genlmsg_reply(reply, info);
2147
2148 exit_err:
2149         nlmsg_free(reply);
2150         return err;
2151 }
2152
2153 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2154 {
2155         struct nlattr **a = info->attrs;
2156         struct nlattr *nla_reply;
2157         struct sk_buff *reply;
2158         struct ovs_header *ovs_reply_header;
2159         struct net *net = sock_net(skb->sk);
2160         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2161         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2162         int err;
2163
2164         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2165                                              &ovs_reply_header);
2166         if (IS_ERR(reply))
2167                 return PTR_ERR(reply);
2168
2169         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2170         if (!nla_reply) {
2171                 err = -EMSGSIZE;
2172                 goto exit_err;
2173         }
2174
2175         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2176                 err = ovs_ct_limit_get_zone_limit(
2177                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2178                         reply);
2179                 if (err)
2180                         goto exit_err;
2181         } else {
2182                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2183                                                       reply);
2184                 if (err)
2185                         goto exit_err;
2186         }
2187
2188         nla_nest_end(reply, nla_reply);
2189         genlmsg_end(reply, ovs_reply_header);
2190         return genlmsg_reply(reply, info);
2191
2192 exit_err:
2193         nlmsg_free(reply);
2194         return err;
2195 }
2196
2197 static struct genl_ops ct_limit_genl_ops[] = {
2198         { .cmd = OVS_CT_LIMIT_CMD_SET,
2199                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2200                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2201                                            * privilege. */
2202                 .doit = ovs_ct_limit_cmd_set,
2203         },
2204         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2205                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2206                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2207                                            * privilege. */
2208                 .doit = ovs_ct_limit_cmd_del,
2209         },
2210         { .cmd = OVS_CT_LIMIT_CMD_GET,
2211                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2212                 .flags = 0,               /* OK for unprivileged users. */
2213                 .doit = ovs_ct_limit_cmd_get,
2214         },
2215 };
2216
2217 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2218         .name = OVS_CT_LIMIT_MCGROUP,
2219 };
2220
2221 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2222         .hdrsize = sizeof(struct ovs_header),
2223         .name = OVS_CT_LIMIT_FAMILY,
2224         .version = OVS_CT_LIMIT_VERSION,
2225         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2226         .policy = ct_limit_policy,
2227         .netnsok = true,
2228         .parallel_ops = true,
2229         .ops = ct_limit_genl_ops,
2230         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2231         .mcgrps = &ovs_ct_limit_multicast_group,
2232         .n_mcgrps = 1,
2233         .module = THIS_MODULE,
2234 };
2235 #endif
2236
2237 int ovs_ct_init(struct net *net)
2238 {
2239         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2240         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2241
2242         if (nf_connlabels_get(net, n_bits - 1)) {
2243                 ovs_net->xt_label = false;
2244                 OVS_NLERR(true, "Failed to set connlabel length");
2245         } else {
2246                 ovs_net->xt_label = true;
2247         }
2248
2249 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2250         return ovs_ct_limit_init(net, ovs_net);
2251 #else
2252         return 0;
2253 #endif
2254 }
2255
2256 void ovs_ct_exit(struct net *net)
2257 {
2258         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2259
2260 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2261         ovs_ct_limit_exit(net, ovs_net);
2262 #endif
2263
2264         if (ovs_net->xt_label)
2265                 nf_connlabels_put(net);
2266 }