net/openvswitch/conntrack.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2015 Nicira, Inc.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/openvswitch.h>
   8 #include <linux/tcp.h>
   9 #include <linux/udp.h>
  10 #include <linux/sctp.h>
  11 #include <linux/static_key.h>
  12 #include <net/ip.h>
  13 #include <net/genetlink.h>
  14 #include <net/netfilter/nf_conntrack_core.h>
  15 #include <net/netfilter/nf_conntrack_count.h>
  16 #include <net/netfilter/nf_conntrack_helper.h>
  17 #include <net/netfilter/nf_conntrack_labels.h>
  18 #include <net/netfilter/nf_conntrack_seqadj.h>
  19 #include <net/netfilter/nf_conntrack_timeout.h>
  20 #include <net/netfilter/nf_conntrack_zones.h>
  21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  22 #include <net/ipv6_frag.h>
  23
  24 #if IS_ENABLED(CONFIG_NF_NAT)
  25 #include <net/netfilter/nf_nat.h>
  26 #endif
  27
  28 #include "datapath.h"
  29 #include "conntrack.h"
  30 #include "flow.h"
  31 #include "flow_netlink.h"
  32
  33 struct ovs_ct_len_tbl {
  34         int maxlen;
  35         int minlen;
  36 };
  37
  38 /* Metadata mark for masked write to conntrack mark */
  39 struct md_mark {
  40         u32 value;
  41         u32 mask;
  42 };
  43
  44 /* Metadata label for masked write to conntrack label. */
  45 struct md_labels {
  46         struct ovs_key_ct_labels value;
  47         struct ovs_key_ct_labels mask;
  48 };
  49
  50 enum ovs_ct_nat {
  51         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  52         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  53         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  54 };
  55
  56 /* Conntrack action context for execution. */
  57 struct ovs_conntrack_info {
  58         struct nf_conntrack_helper *helper;
  59         struct nf_conntrack_zone zone;
  60         struct nf_conn *ct;
  61         u8 commit : 1;
  62         u8 nat : 3;                 /* enum ovs_ct_nat */
  63         u8 force : 1;
  64         u8 have_eventmask : 1;
  65         u16 family;
  66         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  67         struct md_mark mark;
  68         struct md_labels labels;
  69         char timeout[CTNL_TIMEOUT_NAME_MAX];
  70         struct nf_ct_timeout *nf_ct_timeout;
  71 #if IS_ENABLED(CONFIG_NF_NAT)
  72         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  73 #endif
  74 };
  75
  76 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  77 #define OVS_CT_LIMIT_UNLIMITED  0
  78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  79 #define CT_LIMIT_HASH_BUCKETS 512
  80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  81
  82 struct ovs_ct_limit {
  83         /* Elements in ovs_ct_limit_info->limits hash table */
  84         struct hlist_node hlist_node;
  85         struct rcu_head rcu;
  86         u16 zone;
  87         u32 limit;
  88 };
  89
  90 struct ovs_ct_limit_info {
  91         u32 default_limit;
  92         struct hlist_head *limits;
  93         struct nf_conncount_data *data;
  94 };
  95
  96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
  97         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
  98 };
  99 #endif
 100
 101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 102
 103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 104
 105 static u16 key_to_nfproto(const struct sw_flow_key *key)
 106 {
 107         switch (ntohs(key->eth.type)) {
 108         case ETH_P_IP:
 109                 return NFPROTO_IPV4;
 110         case ETH_P_IPV6:
 111                 return NFPROTO_IPV6;
 112         default:
 113                 return NFPROTO_UNSPEC;
 114         }
 115 }
 116
 117 /* Map SKB connection state into the values used by flow definition. */
 118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 119 {
 120         u8 ct_state = OVS_CS_F_TRACKED;
 121
 122         switch (ctinfo) {
 123         case IP_CT_ESTABLISHED_REPLY:
 124         case IP_CT_RELATED_REPLY:
 125                 ct_state |= OVS_CS_F_REPLY_DIR;
 126                 break;
 127         default:
 128                 break;
 129         }
 130
 131         switch (ctinfo) {
 132         case IP_CT_ESTABLISHED:
 133         case IP_CT_ESTABLISHED_REPLY:
 134                 ct_state |= OVS_CS_F_ESTABLISHED;
 135                 break;
 136         case IP_CT_RELATED:
 137         case IP_CT_RELATED_REPLY:
 138                 ct_state |= OVS_CS_F_RELATED;
 139                 break;
 140         case IP_CT_NEW:
 141                 ct_state |= OVS_CS_F_NEW;
 142                 break;
 143         default:
 144                 break;
 145         }
 146
 147         return ct_state;
 148 }
 149
 150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 151 {
 152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 153         return ct ? ct->mark : 0;
 154 #else
 155         return 0;
 156 #endif
 157 }
 158
 159 /* Guard against conntrack labels max size shrinking below 128 bits. */
 160 #if NF_CT_LABELS_MAX_SIZE < 16
 161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 162 #endif
 163
 164 static void ovs_ct_get_labels(const struct nf_conn *ct,
 165                               struct ovs_key_ct_labels *labels)
 166 {
 167         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 168
 169         if (cl)
 170                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 171         else
 172                 memset(labels, 0, OVS_CT_LABELS_LEN);
 173 }
 174
 175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 176                                         const struct nf_conntrack_tuple *orig,
 177                                         u8 icmp_proto)
 178 {
 179         key->ct_orig_proto = orig->dst.protonum;
 180         if (orig->dst.protonum == icmp_proto) {
 181                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 182                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 183         } else {
 184                 key->ct.orig_tp.src = orig->src.u.all;
 185                 key->ct.orig_tp.dst = orig->dst.u.all;
 186         }
 187 }
 188
 189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 190                                 const struct nf_conntrack_zone *zone,
 191                                 const struct nf_conn *ct)
 192 {
 193         key->ct_state = state;
 194         key->ct_zone = zone->id;
 195         key->ct.mark = ovs_ct_get_mark(ct);
 196         ovs_ct_get_labels(ct, &key->ct.labels);
 197
 198         if (ct) {
 199                 const struct nf_conntrack_tuple *orig;
 200
 201                 /* Use the master if we have one. */
 202                 if (ct->master)
 203                         ct = ct->master;
 204                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 205
 206                 /* IP version must match with the master connection. */
 207                 if (key->eth.type == htons(ETH_P_IP) &&
 208                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 209                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 210                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 211                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 212                         return;
 213                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 214                            !sw_flow_key_is_nd(key) &&
 215                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 216                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 217                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 218                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 219                         return;
 220                 }
 221         }
 222         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 223          * original direction key fields.
 224          */
 225         key->ct_orig_proto = 0;
 226 }
 227
 228 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 229  * previously sent the packet to conntrack via the ct action.  If
 230  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 231  * initialized from the connection status.
 232  */
 233 static void ovs_ct_update_key(const struct sk_buff *skb,
 234                               const struct ovs_conntrack_info *info,
 235                               struct sw_flow_key *key, bool post_ct,
 236                               bool keep_nat_flags)
 237 {
 238         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 239         enum ip_conntrack_info ctinfo;
 240         struct nf_conn *ct;
 241         u8 state = 0;
 242
 243         ct = nf_ct_get(skb, &ctinfo);
 244         if (ct) {
 245                 state = ovs_ct_get_state(ctinfo);
 246                 /* All unconfirmed entries are NEW connections. */
 247                 if (!nf_ct_is_confirmed(ct))
 248                         state |= OVS_CS_F_NEW;
 249                 /* OVS persists the related flag for the duration of the
 250                  * connection.
 251                  */
 252                 if (ct->master)
 253                         state |= OVS_CS_F_RELATED;
 254                 if (keep_nat_flags) {
 255                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 256                 } else {
 257                         if (ct->status & IPS_SRC_NAT)
 258                                 state |= OVS_CS_F_SRC_NAT;
 259                         if (ct->status & IPS_DST_NAT)
 260                                 state |= OVS_CS_F_DST_NAT;
 261                 }
 262                 zone = nf_ct_zone(ct);
 263         } else if (post_ct) {
 264                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 265                 if (info)
 266                         zone = &info->zone;
 267         }
 268         __ovs_ct_update_key(key, state, zone, ct);
 269 }
 270
 271 /* This is called to initialize CT key fields possibly coming in from the local
 272  * stack.
 273  */
 274 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 275 {
 276         ovs_ct_update_key(skb, NULL, key, false, false);
 277 }
 278
 279 #define IN6_ADDR_INITIALIZER(ADDR) \
 280         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 281           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 282
 283 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 284                    const struct sw_flow_key *output, struct sk_buff *skb)
 285 {
 286         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 287                 return -EMSGSIZE;
 288
 289         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 290             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 291                 return -EMSGSIZE;
 292
 293         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 294             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 295                 return -EMSGSIZE;
 296
 297         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 298             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 299                     &output->ct.labels))
 300                 return -EMSGSIZE;
 301
 302         if (swkey->ct_orig_proto) {
 303                 if (swkey->eth.type == htons(ETH_P_IP)) {
 304                         struct ovs_key_ct_tuple_ipv4 orig = {
 305                                 output->ipv4.ct_orig.src,
 306                                 output->ipv4.ct_orig.dst,
 307                                 output->ct.orig_tp.src,
 308                                 output->ct.orig_tp.dst,
 309                                 output->ct_orig_proto,
 310                         };
 311                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 312                                     sizeof(orig), &orig))
 313                                 return -EMSGSIZE;
 314                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 315                         struct ovs_key_ct_tuple_ipv6 orig = {
 316                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 317                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 318                                 output->ct.orig_tp.src,
 319                                 output->ct.orig_tp.dst,
 320                                 output->ct_orig_proto,
 321                         };
 322                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 323                                     sizeof(orig), &orig))
 324                                 return -EMSGSIZE;
 325                 }
 326         }
 327
 328         return 0;
 329 }
 330
 331 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 332                            u32 ct_mark, u32 mask)
 333 {
 334 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 335         u32 new_mark;
 336
 337         new_mark = ct_mark | (ct->mark & ~(mask));
 338         if (ct->mark != new_mark) {
 339                 ct->mark = new_mark;
 340                 if (nf_ct_is_confirmed(ct))
 341                         nf_conntrack_event_cache(IPCT_MARK, ct);
 342                 key->ct.mark = new_mark;
 343         }
 344
 345         return 0;
 346 #else
 347         return -ENOTSUPP;
 348 #endif
 349 }
 350
 351 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 352 {
 353         struct nf_conn_labels *cl;
 354
 355         cl = nf_ct_labels_find(ct);
 356         if (!cl) {
 357                 nf_ct_labels_ext_add(ct);
 358                 cl = nf_ct_labels_find(ct);
 359         }
 360
 361         return cl;
 362 }
 363
 364 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 365  * since the new connection is not yet confirmed, and thus no-one else has
 366  * access to it's labels, we simply write them over.
 367  */
 368 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 369                               const struct ovs_key_ct_labels *labels,
 370                               const struct ovs_key_ct_labels *mask)
 371 {
 372         struct nf_conn_labels *cl, *master_cl;
 373         bool have_mask = labels_nonzero(mask);
 374
 375         /* Inherit master's labels to the related connection? */
 376         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 377
 378         if (!master_cl && !have_mask)
 379                 return 0;   /* Nothing to do. */
 380
 381         cl = ovs_ct_get_conn_labels(ct);
 382         if (!cl)
 383                 return -ENOSPC;
 384
 385         /* Inherit the master's labels, if any. */
 386         if (master_cl)
 387                 *cl = *master_cl;
 388
 389         if (have_mask) {
 390                 u32 *dst = (u32 *)cl->bits;
 391                 int i;
 392
 393                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 394                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 395                                 (labels->ct_labels_32[i]
 396                                  & mask->ct_labels_32[i]);
 397         }
 398
 399         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 400          * IPCT_LABEL bit is set in the event cache.
 401          */
 402         nf_conntrack_event_cache(IPCT_LABEL, ct);
 403
 404         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 405
 406         return 0;
 407 }
 408
 409 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 410                              const struct ovs_key_ct_labels *labels,
 411                              const struct ovs_key_ct_labels *mask)
 412 {
 413         struct nf_conn_labels *cl;
 414         int err;
 415
 416         cl = ovs_ct_get_conn_labels(ct);
 417         if (!cl)
 418                 return -ENOSPC;
 419
 420         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 421                                     mask->ct_labels_32,
 422                                     OVS_CT_LABELS_LEN_32);
 423         if (err)
 424                 return err;
 425
 426         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 427
 428         return 0;
 429 }
 430
 431 /* 'skb' should already be pulled to nh_ofs. */
 432 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 433 {
 434         const struct nf_conntrack_helper *helper;
 435         const struct nf_conn_help *help;
 436         enum ip_conntrack_info ctinfo;
 437         unsigned int protoff;
 438         struct nf_conn *ct;
 439         int err;
 440
 441         ct = nf_ct_get(skb, &ctinfo);
 442         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 443                 return NF_ACCEPT;
 444
 445         help = nfct_help(ct);
 446         if (!help)
 447                 return NF_ACCEPT;
 448
 449         helper = rcu_dereference(help->helper);
 450         if (!helper)
 451                 return NF_ACCEPT;
 452
 453         switch (proto) {
 454         case NFPROTO_IPV4:
 455                 protoff = ip_hdrlen(skb);
 456                 break;
 457         case NFPROTO_IPV6: {
 458                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 459                 __be16 frag_off;
 460                 int ofs;
 461
 462                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 463                                        &frag_off);
 464                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 465                         pr_debug("proto header not found\n");
 466                         return NF_ACCEPT;
 467                 }
 468                 protoff = ofs;
 469                 break;
 470         }
 471         default:
 472                 WARN_ONCE(1, "helper invoked on non-IP family!");
 473                 return NF_DROP;
 474         }
 475
 476         err = helper->help(skb, protoff, ct, ctinfo);
 477         if (err != NF_ACCEPT)
 478                 return err;
 479
 480         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 481          * FTP with NAT) adusting the TCP payload size when mangling IP
 482          * addresses and/or port numbers in the text-based control connection.
 483          */
 484         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 485             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 486                 return NF_DROP;
 487         return NF_ACCEPT;
 488 }
 489
 490 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 491  * value if 'skb' is freed.
 492  */
 493 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 494                             u16 zone, struct sk_buff *skb)
 495 {
 496         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 497         int err;
 498
 499         if (key->eth.type == htons(ETH_P_IP)) {
 500                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 501
 502                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 503                 err = ip_defrag(net, skb, user);
 504                 if (err)
 505                         return err;
 506
 507                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 508 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 509         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 510                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 511
 512                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 513                 err = nf_ct_frag6_gather(net, skb, user);
 514                 if (err) {
 515                         if (err != -EINPROGRESS)
 516                                 kfree_skb(skb);
 517                         return err;
 518                 }
 519
 520                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 521                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 522 #endif
 523         } else {
 524                 kfree_skb(skb);
 525                 return -EPFNOSUPPORT;
 526         }
 527
 528         /* The key extracted from the fragment that completed this datagram
 529          * likely didn't have an L4 header, so regenerate it.
 530          */
 531         ovs_flow_key_update_l3l4(skb, key);
 532
 533         key->ip.frag = OVS_FRAG_TYPE_NONE;
 534         skb_clear_hash(skb);
 535         skb->ignore_df = 1;
 536         *OVS_CB(skb) = ovs_cb;
 537
 538         return 0;
 539 }
 540
 541 static struct nf_conntrack_expect *
 542 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 543                    u16 proto, const struct sk_buff *skb)
 544 {
 545         struct nf_conntrack_tuple tuple;
 546         struct nf_conntrack_expect *exp;
 547
 548         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 549                 return NULL;
 550
 551         exp = __nf_ct_expect_find(net, zone, &tuple);
 552         if (exp) {
 553                 struct nf_conntrack_tuple_hash *h;
 554
 555                 /* Delete existing conntrack entry, if it clashes with the
 556                  * expectation.  This can happen since conntrack ALGs do not
 557                  * check for clashes between (new) expectations and existing
 558                  * conntrack entries.  nf_conntrack_in() will check the
 559                  * expectations only if a conntrack entry can not be found,
 560                  * which can lead to OVS finding the expectation (here) in the
 561                  * init direction, but which will not be removed by the
 562                  * nf_conntrack_in() call, if a matching conntrack entry is
 563                  * found instead.  In this case all init direction packets
 564                  * would be reported as new related packets, while reply
 565                  * direction packets would be reported as un-related
 566                  * established packets.
 567                  */
 568                 h = nf_conntrack_find_get(net, zone, &tuple);
 569                 if (h) {
 570                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 571
 572                         nf_ct_delete(ct, 0, 0);
 573                         nf_conntrack_put(&ct->ct_general);
 574                 }
 575         }
 576
 577         return exp;
 578 }
 579
 580 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 581 static enum ip_conntrack_info
 582 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 583 {
 584         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 585
 586         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 587                 return IP_CT_ESTABLISHED_REPLY;
 588         /* Once we've had two way comms, always ESTABLISHED. */
 589         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 590                 return IP_CT_ESTABLISHED;
 591         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 592                 return IP_CT_RELATED;
 593         return IP_CT_NEW;
 594 }
 595
 596 /* Find an existing connection which this packet belongs to without
 597  * re-attributing statistics or modifying the connection state.  This allows an
 598  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 599  *
 600  * Must be called with rcu_read_lock.
 601  *
 602  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 603  * if there is no existing entry.
 604  */
 605 static struct nf_conn *
 606 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 607                      u8 l3num, struct sk_buff *skb, bool natted)
 608 {
 609         struct nf_conntrack_tuple tuple;
 610         struct nf_conntrack_tuple_hash *h;
 611         struct nf_conn *ct;
 612
 613         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 614                                net, &tuple)) {
 615                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 616                 return NULL;
 617         }
 618
 619         /* Must invert the tuple if skb has been transformed by NAT. */
 620         if (natted) {
 621                 struct nf_conntrack_tuple inverse;
 622
 623                 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
 624                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 625                         return NULL;
 626                 }
 627                 tuple = inverse;
 628         }
 629
 630         /* look for tuple match */
 631         h = nf_conntrack_find_get(net, zone, &tuple);
 632         if (!h)
 633                 return NULL;   /* Not found. */
 634
 635         ct = nf_ct_tuplehash_to_ctrack(h);
 636
 637         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 638          * select the other tuplehash to get the right 'ctinfo' bits for this
 639          * packet.
 640          */
 641         if (natted)
 642                 h = &ct->tuplehash[!h->tuple.dst.dir];
 643
 644         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 645         return ct;
 646 }
 647
 648 static
 649 struct nf_conn *ovs_ct_executed(struct net *net,
 650                                 const struct sw_flow_key *key,
 651                                 const struct ovs_conntrack_info *info,
 652                                 struct sk_buff *skb,
 653                                 bool *ct_executed)
 654 {
 655         struct nf_conn *ct = NULL;
 656
 657         /* If no ct, check if we have evidence that an existing conntrack entry
 658          * might be found for this skb.  This happens when we lose a skb->_nfct
 659          * due to an upcall, or if the direction is being forced.  If the
 660          * connection was not confirmed, it is not cached and needs to be run
 661          * through conntrack again.
 662          */
 663         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 664                        !(key->ct_state & OVS_CS_F_INVALID) &&
 665                        (key->ct_zone == info->zone.id);
 666
 667         if (*ct_executed || (!key->ct_state && info->force)) {
 668                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 669                                           !!(key->ct_state &
 670                                           OVS_CS_F_NAT_MASK));
 671         }
 672
 673         return ct;
 674 }
 675
 676 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 677 static bool skb_nfct_cached(struct net *net,
 678                             const struct sw_flow_key *key,
 679                             const struct ovs_conntrack_info *info,
 680                             struct sk_buff *skb)
 681 {
 682         enum ip_conntrack_info ctinfo;
 683         struct nf_conn *ct;
 684         bool ct_executed = true;
 685
 686         ct = nf_ct_get(skb, &ctinfo);
 687         if (!ct)
 688                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 689
 690         if (ct)
 691                 nf_ct_get(skb, &ctinfo);
 692         else
 693                 return false;
 694
 695         if (!net_eq(net, read_pnet(&ct->ct_net)))
 696                 return false;
 697         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 698                 return false;
 699         if (info->helper) {
 700                 struct nf_conn_help *help;
 701
 702                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 703                 if (help && rcu_access_pointer(help->helper) != info->helper)
 704                         return false;
 705         }
 706         if (info->nf_ct_timeout) {
 707                 struct nf_conn_timeout *timeout_ext;
 708
 709                 timeout_ext = nf_ct_timeout_find(ct);
 710                 if (!timeout_ext || info->nf_ct_timeout !=
 711                     rcu_dereference(timeout_ext->timeout))
 712                         return false;
 713         }
 714         /* Force conntrack entry direction to the current packet? */
 715         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 716                 /* Delete the conntrack entry if confirmed, else just release
 717                  * the reference.
 718                  */
 719                 if (nf_ct_is_confirmed(ct))
 720                         nf_ct_delete(ct, 0, 0);
 721
 722                 nf_conntrack_put(&ct->ct_general);
 723                 nf_ct_set(skb, NULL, 0);
 724                 return false;
 725         }
 726
 727         return ct_executed;
 728 }
 729
 730 #if IS_ENABLED(CONFIG_NF_NAT)
 731 /* Modelled after nf_nat_ipv[46]_fn().
 732  * range is only used for new, uninitialized NAT state.
 733  * Returns either NF_ACCEPT or NF_DROP.
 734  */
 735 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 736                               enum ip_conntrack_info ctinfo,
 737                               const struct nf_nat_range2 *range,
 738                               enum nf_nat_manip_type maniptype)
 739 {
 740         int hooknum, nh_off, err = NF_ACCEPT;
 741
 742         nh_off = skb_network_offset(skb);
 743         skb_pull_rcsum(skb, nh_off);
 744
 745         /* See HOOK2MANIP(). */
 746         if (maniptype == NF_NAT_MANIP_SRC)
 747                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 748         else
 749                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 750
 751         switch (ctinfo) {
 752         case IP_CT_RELATED:
 753         case IP_CT_RELATED_REPLY:
 754                 if (IS_ENABLED(CONFIG_NF_NAT) &&
 755                     skb->protocol == htons(ETH_P_IP) &&
 756                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 757                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 758                                                            hooknum))
 759                                 err = NF_DROP;
 760                         goto push;
 761                 } else if (IS_ENABLED(CONFIG_IPV6) &&
 762                            skb->protocol == htons(ETH_P_IPV6)) {
 763                         __be16 frag_off;
 764                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 765                         int hdrlen = ipv6_skip_exthdr(skb,
 766                                                       sizeof(struct ipv6hdr),
 767                                                       &nexthdr, &frag_off);
 768
 769                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 770                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 771                                                                      ctinfo,
 772                                                                      hooknum,
 773                                                                      hdrlen))
 774                                         err = NF_DROP;
 775                                 goto push;
 776                         }
 777                 }
 778                 /* Non-ICMP, fall thru to initialize if needed. */
 779                 /* fall through */
 780         case IP_CT_NEW:
 781                 /* Seen it before?  This can happen for loopback, retrans,
 782                  * or local packets.
 783                  */
 784                 if (!nf_nat_initialized(ct, maniptype)) {
 785                         /* Initialize according to the NAT action. */
 786                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 787                                 /* Action is set up to establish a new
 788                                  * mapping.
 789                                  */
 790                                 ? nf_nat_setup_info(ct, range, maniptype)
 791                                 : nf_nat_alloc_null_binding(ct, hooknum);
 792                         if (err != NF_ACCEPT)
 793                                 goto push;
 794                 }
 795                 break;
 796
 797         case IP_CT_ESTABLISHED:
 798         case IP_CT_ESTABLISHED_REPLY:
 799                 break;
 800
 801         default:
 802                 err = NF_DROP;
 803                 goto push;
 804         }
 805
 806         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 807 push:
 808         skb_push(skb, nh_off);
 809         skb_postpush_rcsum(skb, skb->data, nh_off);
 810
 811         return err;
 812 }
 813
 814 static void ovs_nat_update_key(struct sw_flow_key *key,
 815                                const struct sk_buff *skb,
 816                                enum nf_nat_manip_type maniptype)
 817 {
 818         if (maniptype == NF_NAT_MANIP_SRC) {
 819                 __be16 src;
 820
 821                 key->ct_state |= OVS_CS_F_SRC_NAT;
 822                 if (key->eth.type == htons(ETH_P_IP))
 823                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 824                 else if (key->eth.type == htons(ETH_P_IPV6))
 825                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 826                                sizeof(key->ipv6.addr.src));
 827                 else
 828                         return;
 829
 830                 if (key->ip.proto == IPPROTO_UDP)
 831                         src = udp_hdr(skb)->source;
 832                 else if (key->ip.proto == IPPROTO_TCP)
 833                         src = tcp_hdr(skb)->source;
 834                 else if (key->ip.proto == IPPROTO_SCTP)
 835                         src = sctp_hdr(skb)->source;
 836                 else
 837                         return;
 838
 839                 key->tp.src = src;
 840         } else {
 841                 __be16 dst;
 842
 843                 key->ct_state |= OVS_CS_F_DST_NAT;
 844                 if (key->eth.type == htons(ETH_P_IP))
 845                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 846                 else if (key->eth.type == htons(ETH_P_IPV6))
 847                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 848                                sizeof(key->ipv6.addr.dst));
 849                 else
 850                         return;
 851
 852                 if (key->ip.proto == IPPROTO_UDP)
 853                         dst = udp_hdr(skb)->dest;
 854                 else if (key->ip.proto == IPPROTO_TCP)
 855                         dst = tcp_hdr(skb)->dest;
 856                 else if (key->ip.proto == IPPROTO_SCTP)
 857                         dst = sctp_hdr(skb)->dest;
 858                 else
 859                         return;
 860
 861                 key->tp.dst = dst;
 862         }
 863 }
 864
 865 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 866 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 867                       const struct ovs_conntrack_info *info,
 868                       struct sk_buff *skb, struct nf_conn *ct,
 869                       enum ip_conntrack_info ctinfo)
 870 {
 871         enum nf_nat_manip_type maniptype;
 872         int err;
 873
 874         /* Add NAT extension if not confirmed yet. */
 875         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 876                 return NF_ACCEPT;   /* Can't NAT. */
 877
 878         /* Determine NAT type.
 879          * Check if the NAT type can be deduced from the tracked connection.
 880          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 881          * when committing.
 882          */
 883         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 884             ct->status & IPS_NAT_MASK &&
 885             (ctinfo != IP_CT_RELATED || info->commit)) {
 886                 /* NAT an established or related connection like before. */
 887                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 888                         /* This is the REPLY direction for a connection
 889                          * for which NAT was applied in the forward
 890                          * direction.  Do the reverse NAT.
 891                          */
 892                         maniptype = ct->status & IPS_SRC_NAT
 893                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 894                 else
 895                         maniptype = ct->status & IPS_SRC_NAT
 896                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 897         } else if (info->nat & OVS_CT_SRC_NAT) {
 898                 maniptype = NF_NAT_MANIP_SRC;
 899         } else if (info->nat & OVS_CT_DST_NAT) {
 900                 maniptype = NF_NAT_MANIP_DST;
 901         } else {
 902                 return NF_ACCEPT; /* Connection is not NATed. */
 903         }
 904         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 905
 906         if (err == NF_ACCEPT &&
 907             ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
 908                 if (maniptype == NF_NAT_MANIP_SRC)
 909                         maniptype = NF_NAT_MANIP_DST;
 910                 else
 911                         maniptype = NF_NAT_MANIP_SRC;
 912
 913                 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
 914                                          maniptype);
 915         }
 916
 917         /* Mark NAT done if successful and update the flow key. */
 918         if (err == NF_ACCEPT)
 919                 ovs_nat_update_key(key, skb, maniptype);
 920
 921         return err;
 922 }
 923 #else /* !CONFIG_NF_NAT */
 924 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 925                       const struct ovs_conntrack_info *info,
 926                       struct sk_buff *skb, struct nf_conn *ct,
 927                       enum ip_conntrack_info ctinfo)
 928 {
 929         return NF_ACCEPT;
 930 }
 931 #endif
 932
 933 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 934  * not done already.  Update key with new CT state after passing the packet
 935  * through conntrack.
 936  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 937  * set to NULL and 0 will be returned.
 938  */
 939 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 940                            const struct ovs_conntrack_info *info,
 941                            struct sk_buff *skb)
 942 {
 943         /* If we are recirculating packets to match on conntrack fields and
 944          * committing with a separate conntrack action,  then we don't need to
 945          * actually run the packet through conntrack twice unless it's for a
 946          * different zone.
 947          */
 948         bool cached = skb_nfct_cached(net, key, info, skb);
 949         enum ip_conntrack_info ctinfo;
 950         struct nf_conn *ct;
 951
 952         if (!cached) {
 953                 struct nf_hook_state state = {
 954                         .hook = NF_INET_PRE_ROUTING,
 955                         .pf = info->family,
 956                         .net = net,
 957                 };
 958                 struct nf_conn *tmpl = info->ct;
 959                 int err;
 960
 961                 /* Associate skb with specified zone. */
 962                 if (tmpl) {
 963                         if (skb_nfct(skb))
 964                                 nf_conntrack_put(skb_nfct(skb));
 965                         nf_conntrack_get(&tmpl->ct_general);
 966                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 967                 }
 968
 969                 err = nf_conntrack_in(skb, &state);
 970                 if (err != NF_ACCEPT)
 971                         return -ENOENT;
 972
 973                 /* Clear CT state NAT flags to mark that we have not yet done
 974                  * NAT after the nf_conntrack_in() call.  We can actually clear
 975                  * the whole state, as it will be re-initialized below.
 976                  */
 977                 key->ct_state = 0;
 978
 979                 /* Update the key, but keep the NAT flags. */
 980                 ovs_ct_update_key(skb, info, key, true, true);
 981         }
 982
 983         ct = nf_ct_get(skb, &ctinfo);
 984         if (ct) {
 985                 bool add_helper = false;
 986
 987                 /* Packets starting a new connection must be NATted before the
 988                  * helper, so that the helper knows about the NAT.  We enforce
 989                  * this by delaying both NAT and helper calls for unconfirmed
 990                  * connections until the committing CT action.  For later
 991                  * packets NAT and Helper may be called in either order.
 992                  *
 993                  * NAT will be done only if the CT action has NAT, and only
 994                  * once per packet (per zone), as guarded by the NAT bits in
 995                  * the key->ct_state.
 996                  */
 997                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 998                     (nf_ct_is_confirmed(ct) || info->commit) &&
 999                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1000                         return -EINVAL;
1001                 }
1002
1003                 /* Userspace may decide to perform a ct lookup without a helper
1004                  * specified followed by a (recirculate and) commit with one,
1005                  * or attach a helper in a later commit.  Therefore, for
1006                  * connections which we will commit, we may need to attach
1007                  * the helper here.
1008                  */
1009                 if (info->commit && info->helper && !nfct_help(ct)) {
1010                         int err = __nf_ct_try_assign_helper(ct, info->ct,
1011                                                             GFP_ATOMIC);
1012                         if (err)
1013                                 return err;
1014                         add_helper = true;
1015
1016                         /* helper installed, add seqadj if NAT is required */
1017                         if (info->nat && !nfct_seqadj(ct)) {
1018                                 if (!nfct_seqadj_ext_add(ct))
1019                                         return -EINVAL;
1020                         }
1021                 }
1022
1023                 /* Call the helper only if:
1024                  * - nf_conntrack_in() was executed above ("!cached") or a
1025                  *   helper was just attached ("add_helper") for a confirmed
1026                  *   connection, or
1027                  * - When committing an unconfirmed connection.
1028                  */
1029                 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1030                                               info->commit) &&
1031                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1032                         return -EINVAL;
1033                 }
1034         }
1035
1036         return 0;
1037 }
1038
1039 /* Lookup connection and read fields into key. */
1040 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1041                          const struct ovs_conntrack_info *info,
1042                          struct sk_buff *skb)
1043 {
1044         struct nf_conntrack_expect *exp;
1045
1046         /* If we pass an expected packet through nf_conntrack_in() the
1047          * expectation is typically removed, but the packet could still be
1048          * lost in upcall processing.  To prevent this from happening we
1049          * perform an explicit expectation lookup.  Expected connections are
1050          * always new, and will be passed through conntrack only when they are
1051          * committed, as it is OK to remove the expectation at that time.
1052          */
1053         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1054         if (exp) {
1055                 u8 state;
1056
1057                 /* NOTE: New connections are NATted and Helped only when
1058                  * committed, so we are not calling into NAT here.
1059                  */
1060                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1061                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1062         } else {
1063                 struct nf_conn *ct;
1064                 int err;
1065
1066                 err = __ovs_ct_lookup(net, key, info, skb);
1067                 if (err)
1068                         return err;
1069
1070                 ct = (struct nf_conn *)skb_nfct(skb);
1071                 if (ct)
1072                         nf_ct_deliver_cached_events(ct);
1073         }
1074
1075         return 0;
1076 }
1077
1078 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1079 {
1080         size_t i;
1081
1082         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1083                 if (labels->ct_labels_32[i])
1084                         return true;
1085
1086         return false;
1087 }
1088
1089 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1090 static struct hlist_head *ct_limit_hash_bucket(
1091         const struct ovs_ct_limit_info *info, u16 zone)
1092 {
1093         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1094 }
1095
1096 /* Call with ovs_mutex */
1097 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1098                          struct ovs_ct_limit *new_ct_limit)
1099 {
1100         struct ovs_ct_limit *ct_limit;
1101         struct hlist_head *head;
1102
1103         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1104         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1105                 if (ct_limit->zone == new_ct_limit->zone) {
1106                         hlist_replace_rcu(&ct_limit->hlist_node,
1107                                           &new_ct_limit->hlist_node);
1108                         kfree_rcu(ct_limit, rcu);
1109                         return;
1110                 }
1111         }
1112
1113         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1114 }
1115
1116 /* Call with ovs_mutex */
1117 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1118 {
1119         struct ovs_ct_limit *ct_limit;
1120         struct hlist_head *head;
1121         struct hlist_node *n;
1122
1123         head = ct_limit_hash_bucket(info, zone);
1124         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1125                 if (ct_limit->zone == zone) {
1126                         hlist_del_rcu(&ct_limit->hlist_node);
1127                         kfree_rcu(ct_limit, rcu);
1128                         return;
1129                 }
1130         }
1131 }
1132
1133 /* Call with RCU read lock */
1134 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1135 {
1136         struct ovs_ct_limit *ct_limit;
1137         struct hlist_head *head;
1138
1139         head = ct_limit_hash_bucket(info, zone);
1140         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1141                 if (ct_limit->zone == zone)
1142                         return ct_limit->limit;
1143         }
1144
1145         return info->default_limit;
1146 }
1147
1148 static int ovs_ct_check_limit(struct net *net,
1149                               const struct ovs_conntrack_info *info,
1150                               const struct nf_conntrack_tuple *tuple)
1151 {
1152         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1153         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1154         u32 per_zone_limit, connections;
1155         u32 conncount_key;
1156
1157         conncount_key = info->zone.id;
1158
1159         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1160         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1161                 return 0;
1162
1163         connections = nf_conncount_count(net, ct_limit_info->data,
1164                                          &conncount_key, tuple, &info->zone);
1165         if (connections > per_zone_limit)
1166                 return -ENOMEM;
1167
1168         return 0;
1169 }
1170 #endif
1171
1172 /* Lookup connection and confirm if unconfirmed. */
1173 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1174                          const struct ovs_conntrack_info *info,
1175                          struct sk_buff *skb)
1176 {
1177         enum ip_conntrack_info ctinfo;
1178         struct nf_conn *ct;
1179         int err;
1180
1181         err = __ovs_ct_lookup(net, key, info, skb);
1182         if (err)
1183                 return err;
1184
1185         /* The connection could be invalid, in which case this is a no-op.*/
1186         ct = nf_ct_get(skb, &ctinfo);
1187         if (!ct)
1188                 return 0;
1189
1190 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1191         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1192                 if (!nf_ct_is_confirmed(ct)) {
1193                         err = ovs_ct_check_limit(net, info,
1194                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1195                         if (err) {
1196                                 net_warn_ratelimited("openvswitch: zone: %u "
1197                                         "exceeds conntrack limit\n",
1198                                         info->zone.id);
1199                                 return err;
1200                         }
1201                 }
1202         }
1203 #endif
1204
1205         /* Set the conntrack event mask if given.  NEW and DELETE events have
1206          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1207          * typically would receive many kinds of updates.  Setting the event
1208          * mask allows those events to be filtered.  The set event mask will
1209          * remain in effect for the lifetime of the connection unless changed
1210          * by a further CT action with both the commit flag and the eventmask
1211          * option. */
1212         if (info->have_eventmask) {
1213                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1214
1215                 if (cache)
1216                         cache->ctmask = info->eventmask;
1217         }
1218
1219         /* Apply changes before confirming the connection so that the initial
1220          * conntrack NEW netlink event carries the values given in the CT
1221          * action.
1222          */
1223         if (info->mark.mask) {
1224                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1225                                       info->mark.mask);
1226                 if (err)
1227                         return err;
1228         }
1229         if (!nf_ct_is_confirmed(ct)) {
1230                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1231                                          &info->labels.mask);
1232                 if (err)
1233                         return err;
1234         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1235                    labels_nonzero(&info->labels.mask)) {
1236                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1237                                         &info->labels.mask);
1238                 if (err)
1239                         return err;
1240         }
1241         /* This will take care of sending queued events even if the connection
1242          * is already confirmed.
1243          */
1244         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1245                 return -EINVAL;
1246
1247         return 0;
1248 }
1249
1250 /* Trim the skb to the length specified by the IP/IPv6 header,
1251  * removing any trailing lower-layer padding. This prepares the skb
1252  * for higher-layer processing that assumes skb->len excludes padding
1253  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1254  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1255  */
1256 static int ovs_skb_network_trim(struct sk_buff *skb)
1257 {
1258         unsigned int len;
1259         int err;
1260
1261         switch (skb->protocol) {
1262         case htons(ETH_P_IP):
1263                 len = ntohs(ip_hdr(skb)->tot_len);
1264                 break;
1265         case htons(ETH_P_IPV6):
1266                 len = sizeof(struct ipv6hdr)
1267                         + ntohs(ipv6_hdr(skb)->payload_len);
1268                 break;
1269         default:
1270                 len = skb->len;
1271         }
1272
1273         err = pskb_trim_rcsum(skb, len);
1274         if (err)
1275                 kfree_skb(skb);
1276
1277         return err;
1278 }
1279
1280 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1281  * value if 'skb' is freed.
1282  */
1283 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1284                    struct sw_flow_key *key,
1285                    const struct ovs_conntrack_info *info)
1286 {
1287         int nh_ofs;
1288         int err;
1289
1290         /* The conntrack module expects to be working at L3. */
1291         nh_ofs = skb_network_offset(skb);
1292         skb_pull_rcsum(skb, nh_ofs);
1293
1294         err = ovs_skb_network_trim(skb);
1295         if (err)
1296                 return err;
1297
1298         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1299                 err = handle_fragments(net, key, info->zone.id, skb);
1300                 if (err)
1301                         return err;
1302         }
1303
1304         if (info->commit)
1305                 err = ovs_ct_commit(net, key, info, skb);
1306         else
1307                 err = ovs_ct_lookup(net, key, info, skb);
1308
1309         skb_push(skb, nh_ofs);
1310         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1311         if (err)
1312                 kfree_skb(skb);
1313         return err;
1314 }
1315
1316 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1317 {
1318         if (skb_nfct(skb)) {
1319                 nf_conntrack_put(skb_nfct(skb));
1320                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1321                 ovs_ct_fill_key(skb, key);
1322         }
1323
1324         return 0;
1325 }
1326
1327 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1328                              const struct sw_flow_key *key, bool log)
1329 {
1330         struct nf_conntrack_helper *helper;
1331         struct nf_conn_help *help;
1332         int ret = 0;
1333
1334         helper = nf_conntrack_helper_try_module_get(name, info->family,
1335                                                     key->ip.proto);
1336         if (!helper) {
1337                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1338                 return -EINVAL;
1339         }
1340
1341         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1342         if (!help) {
1343                 nf_conntrack_helper_put(helper);
1344                 return -ENOMEM;
1345         }
1346
1347 #if IS_ENABLED(CONFIG_NF_NAT)
1348         if (info->nat) {
1349                 ret = nf_nat_helper_try_module_get(name, info->family,
1350                                                    key->ip.proto);
1351                 if (ret) {
1352                         nf_conntrack_helper_put(helper);
1353                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1354                                   name, ret);
1355                         return ret;
1356                 }
1357         }
1358 #endif
1359         rcu_assign_pointer(help->helper, helper);
1360         info->helper = helper;
1361         return ret;
1362 }
1363
1364 #if IS_ENABLED(CONFIG_NF_NAT)
1365 static int parse_nat(const struct nlattr *attr,
1366                      struct ovs_conntrack_info *info, bool log)
1367 {
1368         struct nlattr *a;
1369         int rem;
1370         bool have_ip_max = false;
1371         bool have_proto_max = false;
1372         bool ip_vers = (info->family == NFPROTO_IPV6);
1373
1374         nla_for_each_nested(a, attr, rem) {
1375                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1376                         [OVS_NAT_ATTR_SRC] = {0, 0},
1377                         [OVS_NAT_ATTR_DST] = {0, 0},
1378                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1379                                                  sizeof(struct in6_addr)},
1380                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1381                                                  sizeof(struct in6_addr)},
1382                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1383                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1384                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1385                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1386                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1387                 };
1388                 int type = nla_type(a);
1389
1390                 if (type > OVS_NAT_ATTR_MAX) {
1391                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1392                                   type, OVS_NAT_ATTR_MAX);
1393                         return -EINVAL;
1394                 }
1395
1396                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1397                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1398                                   type, nla_len(a),
1399                                   ovs_nat_attr_lens[type][ip_vers]);
1400                         return -EINVAL;
1401                 }
1402
1403                 switch (type) {
1404                 case OVS_NAT_ATTR_SRC:
1405                 case OVS_NAT_ATTR_DST:
1406                         if (info->nat) {
1407                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1408                                 return -ERANGE;
1409                         }
1410                         info->nat |= OVS_CT_NAT;
1411                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1412                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1413                         break;
1414
1415                 case OVS_NAT_ATTR_IP_MIN:
1416                         nla_memcpy(&info->range.min_addr, a,
1417                                    sizeof(info->range.min_addr));
1418                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1419                         break;
1420
1421                 case OVS_NAT_ATTR_IP_MAX:
1422                         have_ip_max = true;
1423                         nla_memcpy(&info->range.max_addr, a,
1424                                    sizeof(info->range.max_addr));
1425                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1426                         break;
1427
1428                 case OVS_NAT_ATTR_PROTO_MIN:
1429                         info->range.min_proto.all = htons(nla_get_u16(a));
1430                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1431                         break;
1432
1433                 case OVS_NAT_ATTR_PROTO_MAX:
1434                         have_proto_max = true;
1435                         info->range.max_proto.all = htons(nla_get_u16(a));
1436                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1437                         break;
1438
1439                 case OVS_NAT_ATTR_PERSISTENT:
1440                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1441                         break;
1442
1443                 case OVS_NAT_ATTR_PROTO_HASH:
1444                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1445                         break;
1446
1447                 case OVS_NAT_ATTR_PROTO_RANDOM:
1448                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1449                         break;
1450
1451                 default:
1452                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1453                         return -EINVAL;
1454                 }
1455         }
1456
1457         if (rem > 0) {
1458                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1459                 return -EINVAL;
1460         }
1461         if (!info->nat) {
1462                 /* Do not allow flags if no type is given. */
1463                 if (info->range.flags) {
1464                         OVS_NLERR(log,
1465                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1466                                   );
1467                         return -EINVAL;
1468                 }
1469                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1470         } else if (!info->commit) {
1471                 OVS_NLERR(log,
1472                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1473                           );
1474                 return -EINVAL;
1475         }
1476         /* Allow missing IP_MAX. */
1477         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1478                 memcpy(&info->range.max_addr, &info->range.min_addr,
1479                        sizeof(info->range.max_addr));
1480         }
1481         /* Allow missing PROTO_MAX. */
1482         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1483             !have_proto_max) {
1484                 info->range.max_proto.all = info->range.min_proto.all;
1485         }
1486         return 0;
1487 }
1488 #endif
1489
1490 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1491         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1492         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1493         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1494                                     .maxlen = sizeof(u16) },
1495         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1496                                     .maxlen = sizeof(struct md_mark) },
1497         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1498                                     .maxlen = sizeof(struct md_labels) },
1499         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1500                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1501 #if IS_ENABLED(CONFIG_NF_NAT)
1502         /* NAT length is checked when parsing the nested attributes. */
1503         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1504 #endif
1505         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1506                                     .maxlen = sizeof(u32) },
1507         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1508                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1509 };
1510
1511 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1512                     const char **helper, bool log)
1513 {
1514         struct nlattr *a;
1515         int rem;
1516
1517         nla_for_each_nested(a, attr, rem) {
1518                 int type = nla_type(a);
1519                 int maxlen;
1520                 int minlen;
1521
1522                 if (type > OVS_CT_ATTR_MAX) {
1523                         OVS_NLERR(log,
1524                                   "Unknown conntrack attr (type=%d, max=%d)",
1525                                   type, OVS_CT_ATTR_MAX);
1526                         return -EINVAL;
1527                 }
1528
1529                 maxlen = ovs_ct_attr_lens[type].maxlen;
1530                 minlen = ovs_ct_attr_lens[type].minlen;
1531                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1532                         OVS_NLERR(log,
1533                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1534                                   type, nla_len(a), maxlen);
1535                         return -EINVAL;
1536                 }
1537
1538                 switch (type) {
1539                 case OVS_CT_ATTR_FORCE_COMMIT:
1540                         info->force = true;
1541                         /* fall through. */
1542                 case OVS_CT_ATTR_COMMIT:
1543                         info->commit = true;
1544                         break;
1545 #ifdef CONFIG_NF_CONNTRACK_ZONES
1546                 case OVS_CT_ATTR_ZONE:
1547                         info->zone.id = nla_get_u16(a);
1548                         break;
1549 #endif
1550 #ifdef CONFIG_NF_CONNTRACK_MARK
1551                 case OVS_CT_ATTR_MARK: {
1552                         struct md_mark *mark = nla_data(a);
1553
1554                         if (!mark->mask) {
1555                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1556                                 return -EINVAL;
1557                         }
1558                         info->mark = *mark;
1559                         break;
1560                 }
1561 #endif
1562 #ifdef CONFIG_NF_CONNTRACK_LABELS
1563                 case OVS_CT_ATTR_LABELS: {
1564                         struct md_labels *labels = nla_data(a);
1565
1566                         if (!labels_nonzero(&labels->mask)) {
1567                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1568                                 return -EINVAL;
1569                         }
1570                         info->labels = *labels;
1571                         break;
1572                 }
1573 #endif
1574                 case OVS_CT_ATTR_HELPER:
1575                         *helper = nla_data(a);
1576                         if (!memchr(*helper, '\0', nla_len(a))) {
1577                                 OVS_NLERR(log, "Invalid conntrack helper");
1578                                 return -EINVAL;
1579                         }
1580                         break;
1581 #if IS_ENABLED(CONFIG_NF_NAT)
1582                 case OVS_CT_ATTR_NAT: {
1583                         int err = parse_nat(a, info, log);
1584
1585                         if (err)
1586                                 return err;
1587                         break;
1588                 }
1589 #endif
1590                 case OVS_CT_ATTR_EVENTMASK:
1591                         info->have_eventmask = true;
1592                         info->eventmask = nla_get_u32(a);
1593                         break;
1594 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1595                 case OVS_CT_ATTR_TIMEOUT:
1596                         memcpy(info->timeout, nla_data(a), nla_len(a));
1597                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1598                                 OVS_NLERR(log, "Invalid conntrack timeout");
1599                                 return -EINVAL;
1600                         }
1601                         break;
1602 #endif
1603
1604                 default:
1605                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1606                                   type);
1607                         return -EINVAL;
1608                 }
1609         }
1610
1611 #ifdef CONFIG_NF_CONNTRACK_MARK
1612         if (!info->commit && info->mark.mask) {
1613                 OVS_NLERR(log,
1614                           "Setting conntrack mark requires 'commit' flag.");
1615                 return -EINVAL;
1616         }
1617 #endif
1618 #ifdef CONFIG_NF_CONNTRACK_LABELS
1619         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1620                 OVS_NLERR(log,
1621                           "Setting conntrack labels requires 'commit' flag.");
1622                 return -EINVAL;
1623         }
1624 #endif
1625         if (rem > 0) {
1626                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1627                 return -EINVAL;
1628         }
1629
1630         return 0;
1631 }
1632
1633 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1634 {
1635         if (attr == OVS_KEY_ATTR_CT_STATE)
1636                 return true;
1637         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1638             attr == OVS_KEY_ATTR_CT_ZONE)
1639                 return true;
1640         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1641             attr == OVS_KEY_ATTR_CT_MARK)
1642                 return true;
1643         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1644             attr == OVS_KEY_ATTR_CT_LABELS) {
1645                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1646
1647                 return ovs_net->xt_label;
1648         }
1649
1650         return false;
1651 }
1652
1653 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1654                        const struct sw_flow_key *key,
1655                        struct sw_flow_actions **sfa,  bool log)
1656 {
1657         struct ovs_conntrack_info ct_info;
1658         const char *helper = NULL;
1659         u16 family;
1660         int err;
1661
1662         family = key_to_nfproto(key);
1663         if (family == NFPROTO_UNSPEC) {
1664                 OVS_NLERR(log, "ct family unspecified");
1665                 return -EINVAL;
1666         }
1667
1668         memset(&ct_info, 0, sizeof(ct_info));
1669         ct_info.family = family;
1670
1671         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1672                         NF_CT_DEFAULT_ZONE_DIR, 0);
1673
1674         err = parse_ct(attr, &ct_info, &helper, log);
1675         if (err)
1676                 return err;
1677
1678         /* Set up template for tracking connections in specific zones. */
1679         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1680         if (!ct_info.ct) {
1681                 OVS_NLERR(log, "Failed to allocate conntrack template");
1682                 return -ENOMEM;
1683         }
1684
1685         if (ct_info.timeout[0]) {
1686                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1687                                       ct_info.timeout))
1688                         pr_info_ratelimited("Failed to associated timeout "
1689                                             "policy `%s'\n", ct_info.timeout);
1690                 else
1691                         ct_info.nf_ct_timeout = rcu_dereference(
1692                                 nf_ct_timeout_find(ct_info.ct)->timeout);
1693
1694         }
1695
1696         if (helper) {
1697                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1698                 if (err)
1699                         goto err_free_ct;
1700         }
1701
1702         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1703                                  sizeof(ct_info), log);
1704         if (err)
1705                 goto err_free_ct;
1706
1707         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1708         nf_conntrack_get(&ct_info.ct->ct_general);
1709         return 0;
1710 err_free_ct:
1711         __ovs_ct_free_action(&ct_info);
1712         return err;
1713 }
1714
1715 #if IS_ENABLED(CONFIG_NF_NAT)
1716 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1717                                struct sk_buff *skb)
1718 {
1719         struct nlattr *start;
1720
1721         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1722         if (!start)
1723                 return false;
1724
1725         if (info->nat & OVS_CT_SRC_NAT) {
1726                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1727                         return false;
1728         } else if (info->nat & OVS_CT_DST_NAT) {
1729                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1730                         return false;
1731         } else {
1732                 goto out;
1733         }
1734
1735         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1736                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1737                     info->family == NFPROTO_IPV4) {
1738                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1739                                             info->range.min_addr.ip) ||
1740                             (info->range.max_addr.ip
1741                              != info->range.min_addr.ip &&
1742                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1743                                               info->range.max_addr.ip))))
1744                                 return false;
1745                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1746                            info->family == NFPROTO_IPV6) {
1747                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1748                                              &info->range.min_addr.in6) ||
1749                             (memcmp(&info->range.max_addr.in6,
1750                                     &info->range.min_addr.in6,
1751                                     sizeof(info->range.max_addr.in6)) &&
1752                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1753                                                &info->range.max_addr.in6))))
1754                                 return false;
1755                 } else {
1756                         return false;
1757                 }
1758         }
1759         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1760             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1761                          ntohs(info->range.min_proto.all)) ||
1762              (info->range.max_proto.all != info->range.min_proto.all &&
1763               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1764                           ntohs(info->range.max_proto.all)))))
1765                 return false;
1766
1767         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1768             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1769                 return false;
1770         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1771             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1772                 return false;
1773         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1774             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1775                 return false;
1776 out:
1777         nla_nest_end(skb, start);
1778
1779         return true;
1780 }
1781 #endif
1782
1783 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1784                           struct sk_buff *skb)
1785 {
1786         struct nlattr *start;
1787
1788         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1789         if (!start)
1790                 return -EMSGSIZE;
1791
1792         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1793                                             ? OVS_CT_ATTR_FORCE_COMMIT
1794                                             : OVS_CT_ATTR_COMMIT))
1795                 return -EMSGSIZE;
1796         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1797             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1798                 return -EMSGSIZE;
1799         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1800             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1801                     &ct_info->mark))
1802                 return -EMSGSIZE;
1803         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1804             labels_nonzero(&ct_info->labels.mask) &&
1805             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1806                     &ct_info->labels))
1807                 return -EMSGSIZE;
1808         if (ct_info->helper) {
1809                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1810                                    ct_info->helper->name))
1811                         return -EMSGSIZE;
1812         }
1813         if (ct_info->have_eventmask &&
1814             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1815                 return -EMSGSIZE;
1816         if (ct_info->timeout[0]) {
1817                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1818                         return -EMSGSIZE;
1819         }
1820
1821 #if IS_ENABLED(CONFIG_NF_NAT)
1822         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1823                 return -EMSGSIZE;
1824 #endif
1825         nla_nest_end(skb, start);
1826
1827         return 0;
1828 }
1829
1830 void ovs_ct_free_action(const struct nlattr *a)
1831 {
1832         struct ovs_conntrack_info *ct_info = nla_data(a);
1833
1834         __ovs_ct_free_action(ct_info);
1835 }
1836
1837 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1838 {
1839         if (ct_info->helper) {
1840 #if IS_ENABLED(CONFIG_NF_NAT)
1841                 if (ct_info->nat)
1842                         nf_nat_helper_put(ct_info->helper);
1843 #endif
1844                 nf_conntrack_helper_put(ct_info->helper);
1845         }
1846         if (ct_info->ct) {
1847                 if (ct_info->timeout[0])
1848                         nf_ct_destroy_timeout(ct_info->ct);
1849                 nf_ct_tmpl_free(ct_info->ct);
1850         }
1851 }
1852
1853 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1854 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1855 {
1856         int i, err;
1857
1858         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1859                                          GFP_KERNEL);
1860         if (!ovs_net->ct_limit_info)
1861                 return -ENOMEM;
1862
1863         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1864         ovs_net->ct_limit_info->limits =
1865                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1866                               GFP_KERNEL);
1867         if (!ovs_net->ct_limit_info->limits) {
1868                 kfree(ovs_net->ct_limit_info);
1869                 return -ENOMEM;
1870         }
1871
1872         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1873                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1874
1875         ovs_net->ct_limit_info->data =
1876                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1877
1878         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1879                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1880                 kfree(ovs_net->ct_limit_info->limits);
1881                 kfree(ovs_net->ct_limit_info);
1882                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1883                 return err;
1884         }
1885         return 0;
1886 }
1887
1888 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1889 {
1890         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1891         int i;
1892
1893         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1894         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1895                 struct hlist_head *head = &info->limits[i];
1896                 struct ovs_ct_limit *ct_limit;
1897
1898                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1899                         kfree_rcu(ct_limit, rcu);
1900         }
1901         kfree(ovs_net->ct_limit_info->limits);
1902         kfree(ovs_net->ct_limit_info);
1903 }
1904
1905 static struct sk_buff *
1906 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1907                              struct ovs_header **ovs_reply_header)
1908 {
1909         struct ovs_header *ovs_header = info->userhdr;
1910         struct sk_buff *skb;
1911
1912         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1913         if (!skb)
1914                 return ERR_PTR(-ENOMEM);
1915
1916         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1917                                         info->snd_seq,
1918                                         &dp_ct_limit_genl_family, 0, cmd);
1919
1920         if (!*ovs_reply_header) {
1921                 nlmsg_free(skb);
1922                 return ERR_PTR(-EMSGSIZE);
1923         }
1924         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1925
1926         return skb;
1927 }
1928
1929 static bool check_zone_id(int zone_id, u16 *pzone)
1930 {
1931         if (zone_id >= 0 && zone_id <= 65535) {
1932                 *pzone = (u16)zone_id;
1933                 return true;
1934         }
1935         return false;
1936 }
1937
1938 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1939                                        struct ovs_ct_limit_info *info)
1940 {
1941         struct ovs_zone_limit *zone_limit;
1942         int rem;
1943         u16 zone;
1944
1945         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1946         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1947
1948         while (rem >= sizeof(*zone_limit)) {
1949                 if (unlikely(zone_limit->zone_id ==
1950                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1951                         ovs_lock();
1952                         info->default_limit = zone_limit->limit;
1953                         ovs_unlock();
1954                 } else if (unlikely(!check_zone_id(
1955                                 zone_limit->zone_id, &zone))) {
1956                         OVS_NLERR(true, "zone id is out of range");
1957                 } else {
1958                         struct ovs_ct_limit *ct_limit;
1959
1960                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1961                         if (!ct_limit)
1962                                 return -ENOMEM;
1963
1964                         ct_limit->zone = zone;
1965                         ct_limit->limit = zone_limit->limit;
1966
1967                         ovs_lock();
1968                         ct_limit_set(info, ct_limit);
1969                         ovs_unlock();
1970                 }
1971                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1972                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1973                                 NLA_ALIGN(sizeof(*zone_limit)));
1974         }
1975
1976         if (rem)
1977                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1978
1979         return 0;
1980 }
1981
1982 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1983                                        struct ovs_ct_limit_info *info)
1984 {
1985         struct ovs_zone_limit *zone_limit;
1986         int rem;
1987         u16 zone;
1988
1989         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1990         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1991
1992         while (rem >= sizeof(*zone_limit)) {
1993                 if (unlikely(zone_limit->zone_id ==
1994                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1995                         ovs_lock();
1996                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
1997                         ovs_unlock();
1998                 } else if (unlikely(!check_zone_id(
1999                                 zone_limit->zone_id, &zone))) {
2000                         OVS_NLERR(true, "zone id is out of range");
2001                 } else {
2002                         ovs_lock();
2003                         ct_limit_del(info, zone);
2004                         ovs_unlock();
2005                 }
2006                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2007                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2008                                 NLA_ALIGN(sizeof(*zone_limit)));
2009         }
2010
2011         if (rem)
2012                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2013
2014         return 0;
2015 }
2016
2017 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2018                                           struct sk_buff *reply)
2019 {
2020         struct ovs_zone_limit zone_limit;
2021         int err;
2022
2023         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2024         zone_limit.limit = info->default_limit;
2025         err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2026         if (err)
2027                 return err;
2028
2029         return 0;
2030 }
2031
2032 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2033                                          struct nf_conncount_data *data,
2034                                          u16 zone_id, u32 limit,
2035                                          struct sk_buff *reply)
2036 {
2037         struct nf_conntrack_zone ct_zone;
2038         struct ovs_zone_limit zone_limit;
2039         u32 conncount_key = zone_id;
2040
2041         zone_limit.zone_id = zone_id;
2042         zone_limit.limit = limit;
2043         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2044
2045         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2046                                               &ct_zone);
2047         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2048 }
2049
2050 static int ovs_ct_limit_get_zone_limit(struct net *net,
2051                                        struct nlattr *nla_zone_limit,
2052                                        struct ovs_ct_limit_info *info,
2053                                        struct sk_buff *reply)
2054 {
2055         struct ovs_zone_limit *zone_limit;
2056         int rem, err;
2057         u32 limit;
2058         u16 zone;
2059
2060         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2061         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2062
2063         while (rem >= sizeof(*zone_limit)) {
2064                 if (unlikely(zone_limit->zone_id ==
2065                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2066                         err = ovs_ct_limit_get_default_limit(info, reply);
2067                         if (err)
2068                                 return err;
2069                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2070                                                         &zone))) {
2071                         OVS_NLERR(true, "zone id is out of range");
2072                 } else {
2073                         rcu_read_lock();
2074                         limit = ct_limit_get(info, zone);
2075                         rcu_read_unlock();
2076
2077                         err = __ovs_ct_limit_get_zone_limit(
2078                                 net, info->data, zone, limit, reply);
2079                         if (err)
2080                                 return err;
2081                 }
2082                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2083                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2084                                 NLA_ALIGN(sizeof(*zone_limit)));
2085         }
2086
2087         if (rem)
2088                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2089
2090         return 0;
2091 }
2092
2093 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2094                                            struct ovs_ct_limit_info *info,
2095                                            struct sk_buff *reply)
2096 {
2097         struct ovs_ct_limit *ct_limit;
2098         struct hlist_head *head;
2099         int i, err = 0;
2100
2101         err = ovs_ct_limit_get_default_limit(info, reply);
2102         if (err)
2103                 return err;
2104
2105         rcu_read_lock();
2106         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2107                 head = &info->limits[i];
2108                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2109                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2110                                 ct_limit->zone, ct_limit->limit, reply);
2111                         if (err)
2112                                 goto exit_err;
2113                 }
2114         }
2115
2116 exit_err:
2117         rcu_read_unlock();
2118         return err;
2119 }
2120
2121 static int ovs_ct_limit_cmd_set(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_SET,
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_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2141                                           ct_limit_info);
2142         if (err)
2143                 goto exit_err;
2144
2145         static_branch_enable(&ovs_ct_limit_enabled);
2146
2147         genlmsg_end(reply, ovs_reply_header);
2148         return genlmsg_reply(reply, info);
2149
2150 exit_err:
2151         nlmsg_free(reply);
2152         return err;
2153 }
2154
2155 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2156 {
2157         struct nlattr **a = info->attrs;
2158         struct sk_buff *reply;
2159         struct ovs_header *ovs_reply_header;
2160         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), 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_DEL,
2165                                              &ovs_reply_header);
2166         if (IS_ERR(reply))
2167                 return PTR_ERR(reply);
2168
2169         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2170                 err = -EINVAL;
2171                 goto exit_err;
2172         }
2173
2174         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2175                                           ct_limit_info);
2176         if (err)
2177                 goto exit_err;
2178
2179         genlmsg_end(reply, ovs_reply_header);
2180         return genlmsg_reply(reply, info);
2181
2182 exit_err:
2183         nlmsg_free(reply);
2184         return err;
2185 }
2186
2187 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2188 {
2189         struct nlattr **a = info->attrs;
2190         struct nlattr *nla_reply;
2191         struct sk_buff *reply;
2192         struct ovs_header *ovs_reply_header;
2193         struct net *net = sock_net(skb->sk);
2194         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2195         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2196         int err;
2197
2198         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2199                                              &ovs_reply_header);
2200         if (IS_ERR(reply))
2201                 return PTR_ERR(reply);
2202
2203         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2204         if (!nla_reply) {
2205                 err = -EMSGSIZE;
2206                 goto exit_err;
2207         }
2208
2209         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2210                 err = ovs_ct_limit_get_zone_limit(
2211                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2212                         reply);
2213                 if (err)
2214                         goto exit_err;
2215         } else {
2216                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2217                                                       reply);
2218                 if (err)
2219                         goto exit_err;
2220         }
2221
2222         nla_nest_end(reply, nla_reply);
2223         genlmsg_end(reply, ovs_reply_header);
2224         return genlmsg_reply(reply, info);
2225
2226 exit_err:
2227         nlmsg_free(reply);
2228         return err;
2229 }
2230
2231 static struct genl_ops ct_limit_genl_ops[] = {
2232         { .cmd = OVS_CT_LIMIT_CMD_SET,
2233                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2234                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2235                                            * privilege. */
2236                 .doit = ovs_ct_limit_cmd_set,
2237         },
2238         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2239                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2240                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2241                                            * privilege. */
2242                 .doit = ovs_ct_limit_cmd_del,
2243         },
2244         { .cmd = OVS_CT_LIMIT_CMD_GET,
2245                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2246                 .flags = 0,               /* OK for unprivileged users. */
2247                 .doit = ovs_ct_limit_cmd_get,
2248         },
2249 };
2250
2251 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2252         .name = OVS_CT_LIMIT_MCGROUP,
2253 };
2254
2255 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2256         .hdrsize = sizeof(struct ovs_header),
2257         .name = OVS_CT_LIMIT_FAMILY,
2258         .version = OVS_CT_LIMIT_VERSION,
2259         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2260         .policy = ct_limit_policy,
2261         .netnsok = true,
2262         .parallel_ops = true,
2263         .ops = ct_limit_genl_ops,
2264         .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2265         .mcgrps = &ovs_ct_limit_multicast_group,
2266         .n_mcgrps = 1,
2267         .module = THIS_MODULE,
2268 };
2269 #endif
2270
2271 int ovs_ct_init(struct net *net)
2272 {
2273         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2274         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2275
2276         if (nf_connlabels_get(net, n_bits - 1)) {
2277                 ovs_net->xt_label = false;
2278                 OVS_NLERR(true, "Failed to set connlabel length");
2279         } else {
2280                 ovs_net->xt_label = true;
2281         }
2282
2283 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2284         return ovs_ct_limit_init(net, ovs_net);
2285 #else
2286         return 0;
2287 #endif
2288 }
2289
2290 void ovs_ct_exit(struct net *net)
2291 {
2292         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2293
2294 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2295         ovs_ct_limit_exit(net, ovs_net);
2296 #endif
2297
2298         if (ovs_net->xt_label)
2299                 nf_connlabels_put(net);
2300 }