net/ipv6/tcp_ipv6.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      TCP over IPv6
   4  *      Linux INET6 implementation
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  *
   9  *      Based on:
  10  *      linux/net/ipv4/tcp.c
  11  *      linux/net/ipv4/tcp_input.c
  12  *      linux/net/ipv4/tcp_output.c
  13  *
  14  *      Fixes:
  15  *      Hideaki YOSHIFUJI       :       sin6_scope_id support
  16  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  17  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
  18  *                                      a single port at the same time.
  19  *      YOSHIFUJI Hideaki @USAGI:       convert /proc/net/tcp6 to seq_file.
  20  */
  21
  22 #include <linux/bottom_half.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/types.h>
  26 #include <linux/socket.h>
  27 #include <linux/sockios.h>
  28 #include <linux/net.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/in.h>
  31 #include <linux/in6.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/init.h>
  34 #include <linux/jhash.h>
  35 #include <linux/ipsec.h>
  36 #include <linux/times.h>
  37 #include <linux/slab.h>
  38 #include <linux/uaccess.h>
  39 #include <linux/ipv6.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/random.h>
  42 #include <linux/indirect_call_wrapper.h>
  43
  44 #include <net/tcp.h>
  45 #include <net/ndisc.h>
  46 #include <net/inet6_hashtables.h>
  47 #include <net/inet6_connection_sock.h>
  48 #include <net/ipv6.h>
  49 #include <net/transp_v6.h>
  50 #include <net/addrconf.h>
  51 #include <net/ip6_route.h>
  52 #include <net/ip6_checksum.h>
  53 #include <net/inet_ecn.h>
  54 #include <net/protocol.h>
  55 #include <net/xfrm.h>
  56 #include <net/snmp.h>
  57 #include <net/dsfield.h>
  58 #include <net/timewait_sock.h>
  59 #include <net/inet_common.h>
  60 #include <net/secure_seq.h>
  61 #include <net/busy_poll.h>
  62
  63 #include <linux/proc_fs.h>
  64 #include <linux/seq_file.h>
  65
  66 #include <crypto/hash.h>
  67 #include <linux/scatterlist.h>
  68
  69 #include <trace/events/tcp.h>
  70
  71 static void     tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
  72 static void     tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
  73                                       struct request_sock *req);
  74
  75 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  76
  77 static const struct inet_connection_sock_af_ops ipv6_mapped;
  78 const struct inet_connection_sock_af_ops ipv6_specific;
  79 #ifdef CONFIG_TCP_MD5SIG
  80 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  81 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  82 #else
  83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
  84                                                    const struct in6_addr *addr,
  85                                                    int l3index)
  86 {
  87         return NULL;
  88 }
  89 #endif
  90
  91 /* Helper returning the inet6 address from a given tcp socket.
  92  * It can be used in TCP stack instead of inet6_sk(sk).
  93  * This avoids a dereference and allow compiler optimizations.
  94  * It is a specialized version of inet6_sk_generic().
  95  */
  96 static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
  97 {
  98         unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
  99
 100         return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
 101 }
 102
 103 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
 104 {
 105         struct dst_entry *dst = skb_dst(skb);
 106
 107         if (dst && dst_hold_safe(dst)) {
 108                 const struct rt6_info *rt = (const struct rt6_info *)dst;
 109
 110                 rcu_assign_pointer(sk->sk_rx_dst, dst);
 111                 sk->sk_rx_dst_ifindex = skb->skb_iif;
 112                 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
 113         }
 114 }
 115
 116 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
 117 {
 118         return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
 119                                 ipv6_hdr(skb)->saddr.s6_addr32,
 120                                 tcp_hdr(skb)->dest,
 121                                 tcp_hdr(skb)->source);
 122 }
 123
 124 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
 125 {
 126         return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
 127                                    ipv6_hdr(skb)->saddr.s6_addr32);
 128 }
 129
 130 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 131                               int addr_len)
 132 {
 133         /* This check is replicated from tcp_v6_connect() and intended to
 134          * prevent BPF program called below from accessing bytes that are out
 135          * of the bound specified by user in addr_len.
 136          */
 137         if (addr_len < SIN6_LEN_RFC2133)
 138                 return -EINVAL;
 139
 140         sock_owned_by_me(sk);
 141
 142         return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
 143 }
 144
 145 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 146                           int addr_len)
 147 {
 148         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 149         struct inet_sock *inet = inet_sk(sk);
 150         struct inet_connection_sock *icsk = inet_csk(sk);
 151         struct ipv6_pinfo *np = tcp_inet6_sk(sk);
 152         struct tcp_sock *tp = tcp_sk(sk);
 153         struct in6_addr *saddr = NULL, *final_p, final;
 154         struct ipv6_txoptions *opt;
 155         struct flowi6 fl6;
 156         struct dst_entry *dst;
 157         int addr_type;
 158         int err;
 159         struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 160
 161         if (addr_len < SIN6_LEN_RFC2133)
 162                 return -EINVAL;
 163
 164         if (usin->sin6_family != AF_INET6)
 165                 return -EAFNOSUPPORT;
 166
 167         memset(&fl6, 0, sizeof(fl6));
 168
 169         if (np->sndflow) {
 170                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 171                 IP6_ECN_flow_init(fl6.flowlabel);
 172                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 173                         struct ip6_flowlabel *flowlabel;
 174                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 175                         if (IS_ERR(flowlabel))
 176                                 return -EINVAL;
 177                         fl6_sock_release(flowlabel);
 178                 }
 179         }
 180
 181         /*
 182          *      connect() to INADDR_ANY means loopback (BSD'ism).
 183          */
 184
 185         if (ipv6_addr_any(&usin->sin6_addr)) {
 186                 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
 187                         ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
 188                                                &usin->sin6_addr);
 189                 else
 190                         usin->sin6_addr = in6addr_loopback;
 191         }
 192
 193         addr_type = ipv6_addr_type(&usin->sin6_addr);
 194
 195         if (addr_type & IPV6_ADDR_MULTICAST)
 196                 return -ENETUNREACH;
 197
 198         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 199                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 200                     usin->sin6_scope_id) {
 201                         /* If interface is set while binding, indices
 202                          * must coincide.
 203                          */
 204                         if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
 205                                 return -EINVAL;
 206
 207                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 208                 }
 209
 210                 /* Connect to link-local address requires an interface */
 211                 if (!sk->sk_bound_dev_if)
 212                         return -EINVAL;
 213         }
 214
 215         if (tp->rx_opt.ts_recent_stamp &&
 216             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 217                 tp->rx_opt.ts_recent = 0;
 218                 tp->rx_opt.ts_recent_stamp = 0;
 219                 WRITE_ONCE(tp->write_seq, 0);
 220         }
 221
 222         sk->sk_v6_daddr = usin->sin6_addr;
 223         np->flow_label = fl6.flowlabel;
 224
 225         /*
 226          *      TCP over IPv4
 227          */
 228
 229         if (addr_type & IPV6_ADDR_MAPPED) {
 230                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 231                 struct sockaddr_in sin;
 232
 233                 if (__ipv6_only_sock(sk))
 234                         return -ENETUNREACH;
 235
 236                 sin.sin_family = AF_INET;
 237                 sin.sin_port = usin->sin6_port;
 238                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 239
 240                 icsk->icsk_af_ops = &ipv6_mapped;
 241                 if (sk_is_mptcp(sk))
 242                         mptcpv6_handle_mapped(sk, true);
 243                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 244 #ifdef CONFIG_TCP_MD5SIG
 245                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 246 #endif
 247
 248                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 249
 250                 if (err) {
 251                         icsk->icsk_ext_hdr_len = exthdrlen;
 252                         icsk->icsk_af_ops = &ipv6_specific;
 253                         if (sk_is_mptcp(sk))
 254                                 mptcpv6_handle_mapped(sk, false);
 255                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 256 #ifdef CONFIG_TCP_MD5SIG
 257                         tp->af_specific = &tcp_sock_ipv6_specific;
 258 #endif
 259                         goto failure;
 260                 }
 261                 np->saddr = sk->sk_v6_rcv_saddr;
 262
 263                 return err;
 264         }
 265
 266         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 267                 saddr = &sk->sk_v6_rcv_saddr;
 268
 269         fl6.flowi6_proto = IPPROTO_TCP;
 270         fl6.daddr = sk->sk_v6_daddr;
 271         fl6.saddr = saddr ? *saddr : np->saddr;
 272         fl6.flowi6_oif = sk->sk_bound_dev_if;
 273         fl6.flowi6_mark = sk->sk_mark;
 274         fl6.fl6_dport = usin->sin6_port;
 275         fl6.fl6_sport = inet->inet_sport;
 276         fl6.flowi6_uid = sk->sk_uid;
 277
 278         opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
 279         final_p = fl6_update_dst(&fl6, opt, &final);
 280
 281         security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
 282
 283         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 284         if (IS_ERR(dst)) {
 285                 err = PTR_ERR(dst);
 286                 goto failure;
 287         }
 288
 289         if (!saddr) {
 290                 saddr = &fl6.saddr;
 291                 sk->sk_v6_rcv_saddr = *saddr;
 292         }
 293
 294         /* set the source address */
 295         np->saddr = *saddr;
 296         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 297
 298         sk->sk_gso_type = SKB_GSO_TCPV6;
 299         ip6_dst_store(sk, dst, NULL, NULL);
 300
 301         icsk->icsk_ext_hdr_len = 0;
 302         if (opt)
 303                 icsk->icsk_ext_hdr_len = opt->opt_flen +
 304                                          opt->opt_nflen;
 305
 306         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 307
 308         inet->inet_dport = usin->sin6_port;
 309
 310         tcp_set_state(sk, TCP_SYN_SENT);
 311         err = inet6_hash_connect(tcp_death_row, sk);
 312         if (err)
 313                 goto late_failure;
 314
 315         sk_set_txhash(sk);
 316
 317         if (likely(!tp->repair)) {
 318                 if (!tp->write_seq)
 319                         WRITE_ONCE(tp->write_seq,
 320                                    secure_tcpv6_seq(np->saddr.s6_addr32,
 321                                                     sk->sk_v6_daddr.s6_addr32,
 322                                                     inet->inet_sport,
 323                                                     inet->inet_dport));
 324                 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
 325                                                    np->saddr.s6_addr32,
 326                                                    sk->sk_v6_daddr.s6_addr32);
 327         }
 328
 329         if (tcp_fastopen_defer_connect(sk, &err))
 330                 return err;
 331         if (err)
 332                 goto late_failure;
 333
 334         err = tcp_connect(sk);
 335         if (err)
 336                 goto late_failure;
 337
 338         return 0;
 339
 340 late_failure:
 341         tcp_set_state(sk, TCP_CLOSE);
 342 failure:
 343         inet->inet_dport = 0;
 344         sk->sk_route_caps = 0;
 345         return err;
 346 }
 347
 348 static void tcp_v6_mtu_reduced(struct sock *sk)
 349 {
 350         struct dst_entry *dst;
 351         u32 mtu;
 352
 353         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 354                 return;
 355
 356         mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
 357
 358         /* Drop requests trying to increase our current mss.
 359          * Check done in __ip6_rt_update_pmtu() is too late.
 360          */
 361         if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
 362                 return;
 363
 364         dst = inet6_csk_update_pmtu(sk, mtu);
 365         if (!dst)
 366                 return;
 367
 368         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 369                 tcp_sync_mss(sk, dst_mtu(dst));
 370                 tcp_simple_retransmit(sk);
 371         }
 372 }
 373
 374 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 375                 u8 type, u8 code, int offset, __be32 info)
 376 {
 377         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 378         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 379         struct net *net = dev_net(skb->dev);
 380         struct request_sock *fastopen;
 381         struct ipv6_pinfo *np;
 382         struct tcp_sock *tp;
 383         __u32 seq, snd_una;
 384         struct sock *sk;
 385         bool fatal;
 386         int err;
 387
 388         sk = __inet6_lookup_established(net, &tcp_hashinfo,
 389                                         &hdr->daddr, th->dest,
 390                                         &hdr->saddr, ntohs(th->source),
 391                                         skb->dev->ifindex, inet6_sdif(skb));
 392
 393         if (!sk) {
 394                 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
 395                                   ICMP6_MIB_INERRORS);
 396                 return -ENOENT;
 397         }
 398
 399         if (sk->sk_state == TCP_TIME_WAIT) {
 400                 inet_twsk_put(inet_twsk(sk));
 401                 return 0;
 402         }
 403         seq = ntohl(th->seq);
 404         fatal = icmpv6_err_convert(type, code, &err);
 405         if (sk->sk_state == TCP_NEW_SYN_RECV) {
 406                 tcp_req_err(sk, seq, fatal);
 407                 return 0;
 408         }
 409
 410         bh_lock_sock(sk);
 411         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 412                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 413
 414         if (sk->sk_state == TCP_CLOSE)
 415                 goto out;
 416
 417         if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
 418                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 419                 goto out;
 420         }
 421
 422         tp = tcp_sk(sk);
 423         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 424         fastopen = rcu_dereference(tp->fastopen_rsk);
 425         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 426         if (sk->sk_state != TCP_LISTEN &&
 427             !between(seq, snd_una, tp->snd_nxt)) {
 428                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 429                 goto out;
 430         }
 431
 432         np = tcp_inet6_sk(sk);
 433
 434         if (type == NDISC_REDIRECT) {
 435                 if (!sock_owned_by_user(sk)) {
 436                         struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 437
 438                         if (dst)
 439                                 dst->ops->redirect(dst, sk, skb);
 440                 }
 441                 goto out;
 442         }
 443
 444         if (type == ICMPV6_PKT_TOOBIG) {
 445                 u32 mtu = ntohl(info);
 446
 447                 /* We are not interested in TCP_LISTEN and open_requests
 448                  * (SYN-ACKs send out by Linux are always <576bytes so
 449                  * they should go through unfragmented).
 450                  */
 451                 if (sk->sk_state == TCP_LISTEN)
 452                         goto out;
 453
 454                 if (!ip6_sk_accept_pmtu(sk))
 455                         goto out;
 456
 457                 if (mtu < IPV6_MIN_MTU)
 458                         goto out;
 459
 460                 WRITE_ONCE(tp->mtu_info, mtu);
 461
 462                 if (!sock_owned_by_user(sk))
 463                         tcp_v6_mtu_reduced(sk);
 464                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 465                                            &sk->sk_tsq_flags))
 466                         sock_hold(sk);
 467                 goto out;
 468         }
 469
 470
 471         /* Might be for an request_sock */
 472         switch (sk->sk_state) {
 473         case TCP_SYN_SENT:
 474         case TCP_SYN_RECV:
 475                 /* Only in fast or simultaneous open. If a fast open socket is
 476                  * already accepted it is treated as a connected one below.
 477                  */
 478                 if (fastopen && !fastopen->sk)
 479                         break;
 480
 481                 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
 482
 483                 if (!sock_owned_by_user(sk)) {
 484                         sk->sk_err = err;
 485                         sk_error_report(sk);            /* Wake people up to see the error (see connect in sock.c) */
 486
 487                         tcp_done(sk);
 488                 } else
 489                         sk->sk_err_soft = err;
 490                 goto out;
 491         case TCP_LISTEN:
 492                 break;
 493         default:
 494                 /* check if this ICMP message allows revert of backoff.
 495                  * (see RFC 6069)
 496                  */
 497                 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
 498                     code == ICMPV6_NOROUTE)
 499                         tcp_ld_RTO_revert(sk, seq);
 500         }
 501
 502         if (!sock_owned_by_user(sk) && np->recverr) {
 503                 sk->sk_err = err;
 504                 sk_error_report(sk);
 505         } else
 506                 sk->sk_err_soft = err;
 507
 508 out:
 509         bh_unlock_sock(sk);
 510         sock_put(sk);
 511         return 0;
 512 }
 513
 514
 515 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 516                               struct flowi *fl,
 517                               struct request_sock *req,
 518                               struct tcp_fastopen_cookie *foc,
 519                               enum tcp_synack_type synack_type,
 520                               struct sk_buff *syn_skb)
 521 {
 522         struct inet_request_sock *ireq = inet_rsk(req);
 523         struct ipv6_pinfo *np = tcp_inet6_sk(sk);
 524         struct ipv6_txoptions *opt;
 525         struct flowi6 *fl6 = &fl->u.ip6;
 526         struct sk_buff *skb;
 527         int err = -ENOMEM;
 528         u8 tclass;
 529
 530         /* First, grab a route. */
 531         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
 532                                                IPPROTO_TCP)) == NULL)
 533                 goto done;
 534
 535         skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
 536
 537         if (skb) {
 538                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 539                                     &ireq->ir_v6_rmt_addr);
 540
 541                 fl6->daddr = ireq->ir_v6_rmt_addr;
 542                 if (np->repflow && ireq->pktopts)
 543                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 544
 545                 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
 546                                 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
 547                                 (np->tclass & INET_ECN_MASK) :
 548                                 np->tclass;
 549
 550                 if (!INET_ECN_is_capable(tclass) &&
 551                     tcp_bpf_ca_needs_ecn((struct sock *)req))
 552                         tclass |= INET_ECN_ECT_0;
 553
 554                 rcu_read_lock();
 555                 opt = ireq->ipv6_opt;
 556                 if (!opt)
 557                         opt = rcu_dereference(np->opt);
 558                 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
 559                                tclass, sk->sk_priority);
 560                 rcu_read_unlock();
 561                 err = net_xmit_eval(err);
 562         }
 563
 564 done:
 565         return err;
 566 }
 567
 568
 569 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 570 {
 571         kfree(inet_rsk(req)->ipv6_opt);
 572         kfree_skb(inet_rsk(req)->pktopts);
 573 }
 574
 575 #ifdef CONFIG_TCP_MD5SIG
 576 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
 577                                                    const struct in6_addr *addr,
 578                                                    int l3index)
 579 {
 580         return tcp_md5_do_lookup(sk, l3index,
 581                                  (union tcp_md5_addr *)addr, AF_INET6);
 582 }
 583
 584 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
 585                                                 const struct sock *addr_sk)
 586 {
 587         int l3index;
 588
 589         l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
 590                                                  addr_sk->sk_bound_dev_if);
 591         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
 592                                     l3index);
 593 }
 594
 595 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
 596                                  sockptr_t optval, int optlen)
 597 {
 598         struct tcp_md5sig cmd;
 599         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 600         int l3index = 0;
 601         u8 prefixlen;
 602         u8 flags;
 603
 604         if (optlen < sizeof(cmd))
 605                 return -EINVAL;
 606
 607         if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
 608                 return -EFAULT;
 609
 610         if (sin6->sin6_family != AF_INET6)
 611                 return -EINVAL;
 612
 613         flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
 614
 615         if (optname == TCP_MD5SIG_EXT &&
 616             cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
 617                 prefixlen = cmd.tcpm_prefixlen;
 618                 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
 619                                         prefixlen > 32))
 620                         return -EINVAL;
 621         } else {
 622                 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
 623         }
 624
 625         if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
 626             cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
 627                 struct net_device *dev;
 628
 629                 rcu_read_lock();
 630                 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
 631                 if (dev && netif_is_l3_master(dev))
 632                         l3index = dev->ifindex;
 633                 rcu_read_unlock();
 634
 635                 /* ok to reference set/not set outside of rcu;
 636                  * right now device MUST be an L3 master
 637                  */
 638                 if (!dev || !l3index)
 639                         return -EINVAL;
 640         }
 641
 642         if (!cmd.tcpm_keylen) {
 643                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 644                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 645                                               AF_INET, prefixlen,
 646                                               l3index, flags);
 647                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 648                                       AF_INET6, prefixlen, l3index, flags);
 649         }
 650
 651         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 652                 return -EINVAL;
 653
 654         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 655                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 656                                       AF_INET, prefixlen, l3index, flags,
 657                                       cmd.tcpm_key, cmd.tcpm_keylen,
 658                                       GFP_KERNEL);
 659
 660         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 661                               AF_INET6, prefixlen, l3index, flags,
 662                               cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 663 }
 664
 665 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
 666                                    const struct in6_addr *daddr,
 667                                    const struct in6_addr *saddr,
 668                                    const struct tcphdr *th, int nbytes)
 669 {
 670         struct tcp6_pseudohdr *bp;
 671         struct scatterlist sg;
 672         struct tcphdr *_th;
 673
 674         bp = hp->scratch;
 675         /* 1. TCP pseudo-header (RFC2460) */
 676         bp->saddr = *saddr;
 677         bp->daddr = *daddr;
 678         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 679         bp->len = cpu_to_be32(nbytes);
 680
 681         _th = (struct tcphdr *)(bp + 1);
 682         memcpy(_th, th, sizeof(*th));
 683         _th->check = 0;
 684
 685         sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
 686         ahash_request_set_crypt(hp->md5_req, &sg, NULL,
 687                                 sizeof(*bp) + sizeof(*th));
 688         return crypto_ahash_update(hp->md5_req);
 689 }
 690
 691 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
 692                                const struct in6_addr *daddr, struct in6_addr *saddr,
 693                                const struct tcphdr *th)
 694 {
 695         struct tcp_md5sig_pool *hp;
 696         struct ahash_request *req;
 697
 698         hp = tcp_get_md5sig_pool();
 699         if (!hp)
 700                 goto clear_hash_noput;
 701         req = hp->md5_req;
 702
 703         if (crypto_ahash_init(req))
 704                 goto clear_hash;
 705         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
 706                 goto clear_hash;
 707         if (tcp_md5_hash_key(hp, key))
 708                 goto clear_hash;
 709         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 710         if (crypto_ahash_final(req))
 711                 goto clear_hash;
 712
 713         tcp_put_md5sig_pool();
 714         return 0;
 715
 716 clear_hash:
 717         tcp_put_md5sig_pool();
 718 clear_hash_noput:
 719         memset(md5_hash, 0, 16);
 720         return 1;
 721 }
 722
 723 static int tcp_v6_md5_hash_skb(char *md5_hash,
 724                                const struct tcp_md5sig_key *key,
 725                                const struct sock *sk,
 726                                const struct sk_buff *skb)
 727 {
 728         const struct in6_addr *saddr, *daddr;
 729         struct tcp_md5sig_pool *hp;
 730         struct ahash_request *req;
 731         const struct tcphdr *th = tcp_hdr(skb);
 732
 733         if (sk) { /* valid for establish/request sockets */
 734                 saddr = &sk->sk_v6_rcv_saddr;
 735                 daddr = &sk->sk_v6_daddr;
 736         } else {
 737                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 738                 saddr = &ip6h->saddr;
 739                 daddr = &ip6h->daddr;
 740         }
 741
 742         hp = tcp_get_md5sig_pool();
 743         if (!hp)
 744                 goto clear_hash_noput;
 745         req = hp->md5_req;
 746
 747         if (crypto_ahash_init(req))
 748                 goto clear_hash;
 749
 750         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
 751                 goto clear_hash;
 752         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 753                 goto clear_hash;
 754         if (tcp_md5_hash_key(hp, key))
 755                 goto clear_hash;
 756         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 757         if (crypto_ahash_final(req))
 758                 goto clear_hash;
 759
 760         tcp_put_md5sig_pool();
 761         return 0;
 762
 763 clear_hash:
 764         tcp_put_md5sig_pool();
 765 clear_hash_noput:
 766         memset(md5_hash, 0, 16);
 767         return 1;
 768 }
 769
 770 #endif
 771
 772 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 773                                     const struct sk_buff *skb,
 774                                     int dif, int sdif)
 775 {
 776 #ifdef CONFIG_TCP_MD5SIG
 777         const __u8 *hash_location = NULL;
 778         struct tcp_md5sig_key *hash_expected;
 779         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 780         const struct tcphdr *th = tcp_hdr(skb);
 781         int genhash, l3index;
 782         u8 newhash[16];
 783
 784         /* sdif set, means packet ingressed via a device
 785          * in an L3 domain and dif is set to the l3mdev
 786          */
 787         l3index = sdif ? dif : 0;
 788
 789         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
 790         hash_location = tcp_parse_md5sig_option(th);
 791
 792         /* We've parsed the options - do we have a hash? */
 793         if (!hash_expected && !hash_location)
 794                 return false;
 795
 796         if (hash_expected && !hash_location) {
 797                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 798                 return true;
 799         }
 800
 801         if (!hash_expected && hash_location) {
 802                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 803                 return true;
 804         }
 805
 806         /* check the signature */
 807         genhash = tcp_v6_md5_hash_skb(newhash,
 808                                       hash_expected,
 809                                       NULL, skb);
 810
 811         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 812                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
 813                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
 814                                      genhash ? "failed" : "mismatch",
 815                                      &ip6h->saddr, ntohs(th->source),
 816                                      &ip6h->daddr, ntohs(th->dest), l3index);
 817                 return true;
 818         }
 819 #endif
 820         return false;
 821 }
 822
 823 static void tcp_v6_init_req(struct request_sock *req,
 824                             const struct sock *sk_listener,
 825                             struct sk_buff *skb)
 826 {
 827         bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
 828         struct inet_request_sock *ireq = inet_rsk(req);
 829         const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
 830
 831         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 832         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 833
 834         /* So that link locals have meaning */
 835         if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
 836             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 837                 ireq->ir_iif = tcp_v6_iif(skb);
 838
 839         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 840             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 841              np->rxopt.bits.rxinfo ||
 842              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 843              np->rxopt.bits.rxohlim || np->repflow)) {
 844                 refcount_inc(&skb->users);
 845                 ireq->pktopts = skb;
 846         }
 847 }
 848
 849 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 850                                           struct sk_buff *skb,
 851                                           struct flowi *fl,
 852                                           struct request_sock *req)
 853 {
 854         tcp_v6_init_req(req, sk, skb);
 855
 856         if (security_inet_conn_request(sk, skb, req))
 857                 return NULL;
 858
 859         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 860 }
 861
 862 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 863         .family         =       AF_INET6,
 864         .obj_size       =       sizeof(struct tcp6_request_sock),
 865         .rtx_syn_ack    =       tcp_rtx_synack,
 866         .send_ack       =       tcp_v6_reqsk_send_ack,
 867         .destructor     =       tcp_v6_reqsk_destructor,
 868         .send_reset     =       tcp_v6_send_reset,
 869         .syn_ack_timeout =      tcp_syn_ack_timeout,
 870 };
 871
 872 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 873         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 874                                 sizeof(struct ipv6hdr),
 875 #ifdef CONFIG_TCP_MD5SIG
 876         .req_md5_lookup =       tcp_v6_md5_lookup,
 877         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 878 #endif
 879 #ifdef CONFIG_SYN_COOKIES
 880         .cookie_init_seq =      cookie_v6_init_sequence,
 881 #endif
 882         .route_req      =       tcp_v6_route_req,
 883         .init_seq       =       tcp_v6_init_seq,
 884         .init_ts_off    =       tcp_v6_init_ts_off,
 885         .send_synack    =       tcp_v6_send_synack,
 886 };
 887
 888 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 889                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 890                                  int oif, struct tcp_md5sig_key *key, int rst,
 891                                  u8 tclass, __be32 label, u32 priority)
 892 {
 893         const struct tcphdr *th = tcp_hdr(skb);
 894         struct tcphdr *t1;
 895         struct sk_buff *buff;
 896         struct flowi6 fl6;
 897         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 898         struct sock *ctl_sk = net->ipv6.tcp_sk;
 899         unsigned int tot_len = sizeof(struct tcphdr);
 900         __be32 mrst = 0, *topt;
 901         struct dst_entry *dst;
 902         __u32 mark = 0;
 903
 904         if (tsecr)
 905                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 906 #ifdef CONFIG_TCP_MD5SIG
 907         if (key)
 908                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 909 #endif
 910
 911 #ifdef CONFIG_MPTCP
 912         if (rst && !key) {
 913                 mrst = mptcp_reset_option(skb);
 914
 915                 if (mrst)
 916                         tot_len += sizeof(__be32);
 917         }
 918 #endif
 919
 920         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 921                          GFP_ATOMIC);
 922         if (!buff)
 923                 return;
 924
 925         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 926
 927         t1 = skb_push(buff, tot_len);
 928         skb_reset_transport_header(buff);
 929
 930         /* Swap the send and the receive. */
 931         memset(t1, 0, sizeof(*t1));
 932         t1->dest = th->source;
 933         t1->source = th->dest;
 934         t1->doff = tot_len / 4;
 935         t1->seq = htonl(seq);
 936         t1->ack_seq = htonl(ack);
 937         t1->ack = !rst || !th->ack;
 938         t1->rst = rst;
 939         t1->window = htons(win);
 940
 941         topt = (__be32 *)(t1 + 1);
 942
 943         if (tsecr) {
 944                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 945                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 946                 *topt++ = htonl(tsval);
 947                 *topt++ = htonl(tsecr);
 948         }
 949
 950         if (mrst)
 951                 *topt++ = mrst;
 952
 953 #ifdef CONFIG_TCP_MD5SIG
 954         if (key) {
 955                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 956                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 957                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 958                                     &ipv6_hdr(skb)->saddr,
 959                                     &ipv6_hdr(skb)->daddr, t1);
 960         }
 961 #endif
 962
 963         memset(&fl6, 0, sizeof(fl6));
 964         fl6.daddr = ipv6_hdr(skb)->saddr;
 965         fl6.saddr = ipv6_hdr(skb)->daddr;
 966         fl6.flowlabel = label;
 967
 968         buff->ip_summed = CHECKSUM_PARTIAL;
 969         buff->csum = 0;
 970
 971         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 972
 973         fl6.flowi6_proto = IPPROTO_TCP;
 974         if (rt6_need_strict(&fl6.daddr) && !oif)
 975                 fl6.flowi6_oif = tcp_v6_iif(skb);
 976         else {
 977                 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
 978                         oif = skb->skb_iif;
 979
 980                 fl6.flowi6_oif = oif;
 981         }
 982
 983         if (sk) {
 984                 if (sk->sk_state == TCP_TIME_WAIT) {
 985                         mark = inet_twsk(sk)->tw_mark;
 986                         /* autoflowlabel relies on buff->hash */
 987                         skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
 988                                      PKT_HASH_TYPE_L4);
 989                 } else {
 990                         mark = sk->sk_mark;
 991                 }
 992                 buff->tstamp = tcp_transmit_time(sk);
 993         }
 994         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
 995         fl6.fl6_dport = t1->dest;
 996         fl6.fl6_sport = t1->source;
 997         fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 998         security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
 999
1000         /* Pass a socket to ip6_dst_lookup either it is for RST
1001          * Underlying function will use this to retrieve the network
1002          * namespace
1003          */
1004         if (sk && sk->sk_state != TCP_TIME_WAIT)
1005                 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
1006         else
1007                 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
1008         if (!IS_ERR(dst)) {
1009                 skb_dst_set(buff, dst);
1010                 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
1011                          tclass & ~INET_ECN_MASK, priority);
1012                 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1013                 if (rst)
1014                         TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1015                 return;
1016         }
1017
1018         kfree_skb(buff);
1019 }
1020
1021 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
1022 {
1023         const struct tcphdr *th = tcp_hdr(skb);
1024         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1025         u32 seq = 0, ack_seq = 0;
1026         struct tcp_md5sig_key *key = NULL;
1027 #ifdef CONFIG_TCP_MD5SIG
1028         const __u8 *hash_location = NULL;
1029         unsigned char newhash[16];
1030         int genhash;
1031         struct sock *sk1 = NULL;
1032 #endif
1033         __be32 label = 0;
1034         u32 priority = 0;
1035         struct net *net;
1036         int oif = 0;
1037
1038         if (th->rst)
1039                 return;
1040
1041         /* If sk not NULL, it means we did a successful lookup and incoming
1042          * route had to be correct. prequeue might have dropped our dst.
1043          */
1044         if (!sk && !ipv6_unicast_destination(skb))
1045                 return;
1046
1047         net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1048 #ifdef CONFIG_TCP_MD5SIG
1049         rcu_read_lock();
1050         hash_location = tcp_parse_md5sig_option(th);
1051         if (sk && sk_fullsock(sk)) {
1052                 int l3index;
1053
1054                 /* sdif set, means packet ingressed via a device
1055                  * in an L3 domain and inet_iif is set to it.
1056                  */
1057                 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1058                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1059         } else if (hash_location) {
1060                 int dif = tcp_v6_iif_l3_slave(skb);
1061                 int sdif = tcp_v6_sdif(skb);
1062                 int l3index;
1063
1064                 /*
1065                  * active side is lost. Try to find listening socket through
1066                  * source port, and then find md5 key through listening socket.
1067                  * we are not loose security here:
1068                  * Incoming packet is checked with md5 hash with finding key,
1069                  * no RST generated if md5 hash doesn't match.
1070                  */
1071                 sk1 = inet6_lookup_listener(net,
1072                                            &tcp_hashinfo, NULL, 0,
1073                                            &ipv6h->saddr,
1074                                            th->source, &ipv6h->daddr,
1075                                            ntohs(th->source), dif, sdif);
1076                 if (!sk1)
1077                         goto out;
1078
1079                 /* sdif set, means packet ingressed via a device
1080                  * in an L3 domain and dif is set to it.
1081                  */
1082                 l3index = tcp_v6_sdif(skb) ? dif : 0;
1083
1084                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1085                 if (!key)
1086                         goto out;
1087
1088                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1089                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1090                         goto out;
1091         }
1092 #endif
1093
1094         if (th->ack)
1095                 seq = ntohl(th->ack_seq);
1096         else
1097                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1098                           (th->doff << 2);
1099
1100         if (sk) {
1101                 oif = sk->sk_bound_dev_if;
1102                 if (sk_fullsock(sk)) {
1103                         const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1104
1105                         trace_tcp_send_reset(sk, skb);
1106                         if (np->repflow)
1107                                 label = ip6_flowlabel(ipv6h);
1108                         priority = sk->sk_priority;
1109                 }
1110                 if (sk->sk_state == TCP_TIME_WAIT) {
1111                         label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1112                         priority = inet_twsk(sk)->tw_priority;
1113                 }
1114         } else {
1115                 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1116                         label = ip6_flowlabel(ipv6h);
1117         }
1118
1119         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1120                              ipv6_get_dsfield(ipv6h), label, priority);
1121
1122 #ifdef CONFIG_TCP_MD5SIG
1123 out:
1124         rcu_read_unlock();
1125 #endif
1126 }
1127
1128 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1129                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1130                             struct tcp_md5sig_key *key, u8 tclass,
1131                             __be32 label, u32 priority)
1132 {
1133         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1134                              tclass, label, priority);
1135 }
1136
1137 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1138 {
1139         struct inet_timewait_sock *tw = inet_twsk(sk);
1140         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1141
1142         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1143                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1144                         tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1145                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1146                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1147
1148         inet_twsk_put(tw);
1149 }
1150
1151 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1152                                   struct request_sock *req)
1153 {
1154         int l3index;
1155
1156         l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1157
1158         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1159          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1160          */
1161         /* RFC 7323 2.3
1162          * The window field (SEG.WND) of every outgoing segment, with the
1163          * exception of <SYN> segments, MUST be right-shifted by
1164          * Rcv.Wind.Shift bits:
1165          */
1166         tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1167                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1168                         tcp_rsk(req)->rcv_nxt,
1169                         req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1170                         tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1171                         req->ts_recent, sk->sk_bound_dev_if,
1172                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1173                         ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority);
1174 }
1175
1176
1177 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1178 {
1179 #ifdef CONFIG_SYN_COOKIES
1180         const struct tcphdr *th = tcp_hdr(skb);
1181
1182         if (!th->syn)
1183                 sk = cookie_v6_check(sk, skb);
1184 #endif
1185         return sk;
1186 }
1187
1188 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1189                          struct tcphdr *th, u32 *cookie)
1190 {
1191         u16 mss = 0;
1192 #ifdef CONFIG_SYN_COOKIES
1193         mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1194                                     &tcp_request_sock_ipv6_ops, sk, th);
1195         if (mss) {
1196                 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1197                 tcp_synq_overflow(sk);
1198         }
1199 #endif
1200         return mss;
1201 }
1202
1203 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1204 {
1205         if (skb->protocol == htons(ETH_P_IP))
1206                 return tcp_v4_conn_request(sk, skb);
1207
1208         if (!ipv6_unicast_destination(skb))
1209                 goto drop;
1210
1211         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1212                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1213                 return 0;
1214         }
1215
1216         return tcp_conn_request(&tcp6_request_sock_ops,
1217                                 &tcp_request_sock_ipv6_ops, sk, skb);
1218
1219 drop:
1220         tcp_listendrop(sk);
1221         return 0; /* don't send reset */
1222 }
1223
1224 static void tcp_v6_restore_cb(struct sk_buff *skb)
1225 {
1226         /* We need to move header back to the beginning if xfrm6_policy_check()
1227          * and tcp_v6_fill_cb() are going to be called again.
1228          * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1229          */
1230         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1231                 sizeof(struct inet6_skb_parm));
1232 }
1233
1234 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1235                                          struct request_sock *req,
1236                                          struct dst_entry *dst,
1237                                          struct request_sock *req_unhash,
1238                                          bool *own_req)
1239 {
1240         struct inet_request_sock *ireq;
1241         struct ipv6_pinfo *newnp;
1242         const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1243         struct ipv6_txoptions *opt;
1244         struct inet_sock *newinet;
1245         bool found_dup_sk = false;
1246         struct tcp_sock *newtp;
1247         struct sock *newsk;
1248 #ifdef CONFIG_TCP_MD5SIG
1249         struct tcp_md5sig_key *key;
1250         int l3index;
1251 #endif
1252         struct flowi6 fl6;
1253
1254         if (skb->protocol == htons(ETH_P_IP)) {
1255                 /*
1256                  *      v6 mapped
1257                  */
1258
1259                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1260                                              req_unhash, own_req);
1261
1262                 if (!newsk)
1263                         return NULL;
1264
1265                 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1266
1267                 newinet = inet_sk(newsk);
1268                 newnp = tcp_inet6_sk(newsk);
1269                 newtp = tcp_sk(newsk);
1270
1271                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1272
1273                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1274
1275                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1276                 if (sk_is_mptcp(newsk))
1277                         mptcpv6_handle_mapped(newsk, true);
1278                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1279 #ifdef CONFIG_TCP_MD5SIG
1280                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1281 #endif
1282
1283                 newnp->ipv6_mc_list = NULL;
1284                 newnp->ipv6_ac_list = NULL;
1285                 newnp->ipv6_fl_list = NULL;
1286                 newnp->pktoptions  = NULL;
1287                 newnp->opt         = NULL;
1288                 newnp->mcast_oif   = inet_iif(skb);
1289                 newnp->mcast_hops  = ip_hdr(skb)->ttl;
1290                 newnp->rcv_flowinfo = 0;
1291                 if (np->repflow)
1292                         newnp->flow_label = 0;
1293
1294                 /*
1295                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1296                  * here, tcp_create_openreq_child now does this for us, see the comment in
1297                  * that function for the gory details. -acme
1298                  */
1299
1300                 /* It is tricky place. Until this moment IPv4 tcp
1301                    worked with IPv6 icsk.icsk_af_ops.
1302                    Sync it now.
1303                  */
1304                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1305
1306                 return newsk;
1307         }
1308
1309         ireq = inet_rsk(req);
1310
1311         if (sk_acceptq_is_full(sk))
1312                 goto out_overflow;
1313
1314         if (!dst) {
1315                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1316                 if (!dst)
1317                         goto out;
1318         }
1319
1320         newsk = tcp_create_openreq_child(sk, req, skb);
1321         if (!newsk)
1322                 goto out_nonewsk;
1323
1324         /*
1325          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1326          * count here, tcp_create_openreq_child now does this for us, see the
1327          * comment in that function for the gory details. -acme
1328          */
1329
1330         newsk->sk_gso_type = SKB_GSO_TCPV6;
1331         ip6_dst_store(newsk, dst, NULL, NULL);
1332         inet6_sk_rx_dst_set(newsk, skb);
1333
1334         inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1335
1336         newtp = tcp_sk(newsk);
1337         newinet = inet_sk(newsk);
1338         newnp = tcp_inet6_sk(newsk);
1339
1340         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1341
1342         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1343         newnp->saddr = ireq->ir_v6_loc_addr;
1344         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1345         newsk->sk_bound_dev_if = ireq->ir_iif;
1346
1347         /* Now IPv6 options...
1348
1349            First: no IPv4 options.
1350          */
1351         newinet->inet_opt = NULL;
1352         newnp->ipv6_mc_list = NULL;
1353         newnp->ipv6_ac_list = NULL;
1354         newnp->ipv6_fl_list = NULL;
1355
1356         /* Clone RX bits */
1357         newnp->rxopt.all = np->rxopt.all;
1358
1359         newnp->pktoptions = NULL;
1360         newnp->opt        = NULL;
1361         newnp->mcast_oif  = tcp_v6_iif(skb);
1362         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1363         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1364         if (np->repflow)
1365                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1366
1367         /* Set ToS of the new socket based upon the value of incoming SYN.
1368          * ECT bits are set later in tcp_init_transfer().
1369          */
1370         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1371                 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1372
1373         /* Clone native IPv6 options from listening socket (if any)
1374
1375            Yes, keeping reference count would be much more clever,
1376            but we make one more one thing there: reattach optmem
1377            to newsk.
1378          */
1379         opt = ireq->ipv6_opt;
1380         if (!opt)
1381                 opt = rcu_dereference(np->opt);
1382         if (opt) {
1383                 opt = ipv6_dup_options(newsk, opt);
1384                 RCU_INIT_POINTER(newnp->opt, opt);
1385         }
1386         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1387         if (opt)
1388                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1389                                                     opt->opt_flen;
1390
1391         tcp_ca_openreq_child(newsk, dst);
1392
1393         tcp_sync_mss(newsk, dst_mtu(dst));
1394         newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1395
1396         tcp_initialize_rcv_mss(newsk);
1397
1398         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1399         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1400
1401 #ifdef CONFIG_TCP_MD5SIG
1402         l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1403
1404         /* Copy over the MD5 key from the original socket */
1405         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1406         if (key) {
1407                 /* We're using one, so create a matching key
1408                  * on the newsk structure. If we fail to get
1409                  * memory, then we end up not copying the key
1410                  * across. Shucks.
1411                  */
1412                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1413                                AF_INET6, 128, l3index, key->flags, key->key, key->keylen,
1414                                sk_gfp_mask(sk, GFP_ATOMIC));
1415         }
1416 #endif
1417
1418         if (__inet_inherit_port(sk, newsk) < 0) {
1419                 inet_csk_prepare_forced_close(newsk);
1420                 tcp_done(newsk);
1421                 goto out;
1422         }
1423         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1424                                        &found_dup_sk);
1425         if (*own_req) {
1426                 tcp_move_syn(newtp, req);
1427
1428                 /* Clone pktoptions received with SYN, if we own the req */
1429                 if (ireq->pktopts) {
1430                         newnp->pktoptions = skb_clone(ireq->pktopts,
1431                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1432                         consume_skb(ireq->pktopts);
1433                         ireq->pktopts = NULL;
1434                         if (newnp->pktoptions) {
1435                                 tcp_v6_restore_cb(newnp->pktoptions);
1436                                 skb_set_owner_r(newnp->pktoptions, newsk);
1437                         }
1438                 }
1439         } else {
1440                 if (!req_unhash && found_dup_sk) {
1441                         /* This code path should only be executed in the
1442                          * syncookie case only
1443                          */
1444                         bh_unlock_sock(newsk);
1445                         sock_put(newsk);
1446                         newsk = NULL;
1447                 }
1448         }
1449
1450         return newsk;
1451
1452 out_overflow:
1453         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1454 out_nonewsk:
1455         dst_release(dst);
1456 out:
1457         tcp_listendrop(sk);
1458         return NULL;
1459 }
1460
1461 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1462                                                            u32));
1463 /* The socket must have it's spinlock held when we get
1464  * here, unless it is a TCP_LISTEN socket.
1465  *
1466  * We have a potential double-lock case here, so even when
1467  * doing backlog processing we use the BH locking scheme.
1468  * This is because we cannot sleep with the original spinlock
1469  * held.
1470  */
1471 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1472 {
1473         struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1474         struct sk_buff *opt_skb = NULL;
1475         struct tcp_sock *tp;
1476
1477         /* Imagine: socket is IPv6. IPv4 packet arrives,
1478            goes to IPv4 receive handler and backlogged.
1479            From backlog it always goes here. Kerboom...
1480            Fortunately, tcp_rcv_established and rcv_established
1481            handle them correctly, but it is not case with
1482            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1483          */
1484
1485         if (skb->protocol == htons(ETH_P_IP))
1486                 return tcp_v4_do_rcv(sk, skb);
1487
1488         /*
1489          *      socket locking is here for SMP purposes as backlog rcv
1490          *      is currently called with bh processing disabled.
1491          */
1492
1493         /* Do Stevens' IPV6_PKTOPTIONS.
1494
1495            Yes, guys, it is the only place in our code, where we
1496            may make it not affecting IPv4.
1497            The rest of code is protocol independent,
1498            and I do not like idea to uglify IPv4.
1499
1500            Actually, all the idea behind IPV6_PKTOPTIONS
1501            looks not very well thought. For now we latch
1502            options, received in the last packet, enqueued
1503            by tcp. Feel free to propose better solution.
1504                                                --ANK (980728)
1505          */
1506         if (np->rxopt.all)
1507                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1508
1509         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1510                 struct dst_entry *dst;
1511
1512                 dst = rcu_dereference_protected(sk->sk_rx_dst,
1513                                                 lockdep_sock_is_held(sk));
1514
1515                 sock_rps_save_rxhash(sk, skb);
1516                 sk_mark_napi_id(sk, skb);
1517                 if (dst) {
1518                         if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1519                             INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1520                                             dst, sk->sk_rx_dst_cookie) == NULL) {
1521                                 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1522                                 dst_release(dst);
1523                         }
1524                 }
1525
1526                 tcp_rcv_established(sk, skb);
1527                 if (opt_skb)
1528                         goto ipv6_pktoptions;
1529                 return 0;
1530         }
1531
1532         if (tcp_checksum_complete(skb))
1533                 goto csum_err;
1534
1535         if (sk->sk_state == TCP_LISTEN) {
1536                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1537
1538                 if (!nsk)
1539                         goto discard;
1540
1541                 if (nsk != sk) {
1542                         if (tcp_child_process(sk, nsk, skb))
1543                                 goto reset;
1544                         if (opt_skb)
1545                                 __kfree_skb(opt_skb);
1546                         return 0;
1547                 }
1548         } else
1549                 sock_rps_save_rxhash(sk, skb);
1550
1551         if (tcp_rcv_state_process(sk, skb))
1552                 goto reset;
1553         if (opt_skb)
1554                 goto ipv6_pktoptions;
1555         return 0;
1556
1557 reset:
1558         tcp_v6_send_reset(sk, skb);
1559 discard:
1560         if (opt_skb)
1561                 __kfree_skb(opt_skb);
1562         kfree_skb(skb);
1563         return 0;
1564 csum_err:
1565         trace_tcp_bad_csum(skb);
1566         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1567         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1568         goto discard;
1569
1570
1571 ipv6_pktoptions:
1572         /* Do you ask, what is it?
1573
1574            1. skb was enqueued by tcp.
1575            2. skb is added to tail of read queue, rather than out of order.
1576            3. socket is not in passive state.
1577            4. Finally, it really contains options, which user wants to receive.
1578          */
1579         tp = tcp_sk(sk);
1580         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1581             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1582                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1583                         np->mcast_oif = tcp_v6_iif(opt_skb);
1584                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1585                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1586                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1587                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1588                 if (np->repflow)
1589                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1590                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1591                         skb_set_owner_r(opt_skb, sk);
1592                         tcp_v6_restore_cb(opt_skb);
1593                         opt_skb = xchg(&np->pktoptions, opt_skb);
1594                 } else {
1595                         __kfree_skb(opt_skb);
1596                         opt_skb = xchg(&np->pktoptions, NULL);
1597                 }
1598         }
1599
1600         kfree_skb(opt_skb);
1601         return 0;
1602 }
1603
1604 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1605                            const struct tcphdr *th)
1606 {
1607         /* This is tricky: we move IP6CB at its correct location into
1608          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1609          * _decode_session6() uses IP6CB().
1610          * barrier() makes sure compiler won't play aliasing games.
1611          */
1612         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1613                 sizeof(struct inet6_skb_parm));
1614         barrier();
1615
1616         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1617         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1618                                     skb->len - th->doff*4);
1619         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1620         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1621         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1622         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1623         TCP_SKB_CB(skb)->sacked = 0;
1624         TCP_SKB_CB(skb)->has_rxtstamp =
1625                         skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1626 }
1627
1628 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1629 {
1630         struct sk_buff *skb_to_free;
1631         int sdif = inet6_sdif(skb);
1632         int dif = inet6_iif(skb);
1633         const struct tcphdr *th;
1634         const struct ipv6hdr *hdr;
1635         bool refcounted;
1636         struct sock *sk;
1637         int ret;
1638         struct net *net = dev_net(skb->dev);
1639
1640         if (skb->pkt_type != PACKET_HOST)
1641                 goto discard_it;
1642
1643         /*
1644          *      Count it even if it's bad.
1645          */
1646         __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1647
1648         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1649                 goto discard_it;
1650
1651         th = (const struct tcphdr *)skb->data;
1652
1653         if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1654                 goto bad_packet;
1655         if (!pskb_may_pull(skb, th->doff*4))
1656                 goto discard_it;
1657
1658         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1659                 goto csum_error;
1660
1661         th = (const struct tcphdr *)skb->data;
1662         hdr = ipv6_hdr(skb);
1663
1664 lookup:
1665         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1666                                 th->source, th->dest, inet6_iif(skb), sdif,
1667                                 &refcounted);
1668         if (!sk)
1669                 goto no_tcp_socket;
1670
1671 process:
1672         if (sk->sk_state == TCP_TIME_WAIT)
1673                 goto do_time_wait;
1674
1675         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1676                 struct request_sock *req = inet_reqsk(sk);
1677                 bool req_stolen = false;
1678                 struct sock *nsk;
1679
1680                 sk = req->rsk_listener;
1681                 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1682                         sk_drops_add(sk, skb);
1683                         reqsk_put(req);
1684                         goto discard_it;
1685                 }
1686                 if (tcp_checksum_complete(skb)) {
1687                         reqsk_put(req);
1688                         goto csum_error;
1689                 }
1690                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1691                         nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1692                         if (!nsk) {
1693                                 inet_csk_reqsk_queue_drop_and_put(sk, req);
1694                                 goto lookup;
1695                         }
1696                         sk = nsk;
1697                         /* reuseport_migrate_sock() has already held one sk_refcnt
1698                          * before returning.
1699                          */
1700                 } else {
1701                         sock_hold(sk);
1702                 }
1703                 refcounted = true;
1704                 nsk = NULL;
1705                 if (!tcp_filter(sk, skb)) {
1706                         th = (const struct tcphdr *)skb->data;
1707                         hdr = ipv6_hdr(skb);
1708                         tcp_v6_fill_cb(skb, hdr, th);
1709                         nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1710                 }
1711                 if (!nsk) {
1712                         reqsk_put(req);
1713                         if (req_stolen) {
1714                                 /* Another cpu got exclusive access to req
1715                                  * and created a full blown socket.
1716                                  * Try to feed this packet to this socket
1717                                  * instead of discarding it.
1718                                  */
1719                                 tcp_v6_restore_cb(skb);
1720                                 sock_put(sk);
1721                                 goto lookup;
1722                         }
1723                         goto discard_and_relse;
1724                 }
1725                 if (nsk == sk) {
1726                         reqsk_put(req);
1727                         tcp_v6_restore_cb(skb);
1728                 } else if (tcp_child_process(sk, nsk, skb)) {
1729                         tcp_v6_send_reset(nsk, skb);
1730                         goto discard_and_relse;
1731                 } else {
1732                         sock_put(sk);
1733                         return 0;
1734                 }
1735         }
1736         if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1737                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1738                 goto discard_and_relse;
1739         }
1740
1741         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1742                 goto discard_and_relse;
1743
1744         if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1745                 goto discard_and_relse;
1746
1747         if (tcp_filter(sk, skb))
1748                 goto discard_and_relse;
1749         th = (const struct tcphdr *)skb->data;
1750         hdr = ipv6_hdr(skb);
1751         tcp_v6_fill_cb(skb, hdr, th);
1752
1753         skb->dev = NULL;
1754
1755         if (sk->sk_state == TCP_LISTEN) {
1756                 ret = tcp_v6_do_rcv(sk, skb);
1757                 goto put_and_return;
1758         }
1759
1760         sk_incoming_cpu_update(sk);
1761
1762         bh_lock_sock_nested(sk);
1763         tcp_segs_in(tcp_sk(sk), skb);
1764         ret = 0;
1765         if (!sock_owned_by_user(sk)) {
1766                 skb_to_free = sk->sk_rx_skb_cache;
1767                 sk->sk_rx_skb_cache = NULL;
1768                 ret = tcp_v6_do_rcv(sk, skb);
1769         } else {
1770                 if (tcp_add_backlog(sk, skb))
1771                         goto discard_and_relse;
1772                 skb_to_free = NULL;
1773         }
1774         bh_unlock_sock(sk);
1775         if (skb_to_free)
1776                 __kfree_skb(skb_to_free);
1777 put_and_return:
1778         if (refcounted)
1779                 sock_put(sk);
1780         return ret ? -1 : 0;
1781
1782 no_tcp_socket:
1783         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1784                 goto discard_it;
1785
1786         tcp_v6_fill_cb(skb, hdr, th);
1787
1788         if (tcp_checksum_complete(skb)) {
1789 csum_error:
1790                 trace_tcp_bad_csum(skb);
1791                 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1792 bad_packet:
1793                 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1794         } else {
1795                 tcp_v6_send_reset(NULL, skb);
1796         }
1797
1798 discard_it:
1799         kfree_skb(skb);
1800         return 0;
1801
1802 discard_and_relse:
1803         sk_drops_add(sk, skb);
1804         if (refcounted)
1805                 sock_put(sk);
1806         goto discard_it;
1807
1808 do_time_wait:
1809         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1810                 inet_twsk_put(inet_twsk(sk));
1811                 goto discard_it;
1812         }
1813
1814         tcp_v6_fill_cb(skb, hdr, th);
1815
1816         if (tcp_checksum_complete(skb)) {
1817                 inet_twsk_put(inet_twsk(sk));
1818                 goto csum_error;
1819         }
1820
1821         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1822         case TCP_TW_SYN:
1823         {
1824                 struct sock *sk2;
1825
1826                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1827                                             skb, __tcp_hdrlen(th),
1828                                             &ipv6_hdr(skb)->saddr, th->source,
1829                                             &ipv6_hdr(skb)->daddr,
1830                                             ntohs(th->dest),
1831                                             tcp_v6_iif_l3_slave(skb),
1832                                             sdif);
1833                 if (sk2) {
1834                         struct inet_timewait_sock *tw = inet_twsk(sk);
1835                         inet_twsk_deschedule_put(tw);
1836                         sk = sk2;
1837                         tcp_v6_restore_cb(skb);
1838                         refcounted = false;
1839                         goto process;
1840                 }
1841         }
1842                 /* to ACK */
1843                 fallthrough;
1844         case TCP_TW_ACK:
1845                 tcp_v6_timewait_ack(sk, skb);
1846                 break;
1847         case TCP_TW_RST:
1848                 tcp_v6_send_reset(sk, skb);
1849                 inet_twsk_deschedule_put(inet_twsk(sk));
1850                 goto discard_it;
1851         case TCP_TW_SUCCESS:
1852                 ;
1853         }
1854         goto discard_it;
1855 }
1856
1857 void tcp_v6_early_demux(struct sk_buff *skb)
1858 {
1859         const struct ipv6hdr *hdr;
1860         const struct tcphdr *th;
1861         struct sock *sk;
1862
1863         if (skb->pkt_type != PACKET_HOST)
1864                 return;
1865
1866         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1867                 return;
1868
1869         hdr = ipv6_hdr(skb);
1870         th = tcp_hdr(skb);
1871
1872         if (th->doff < sizeof(struct tcphdr) / 4)
1873                 return;
1874
1875         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1876         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1877                                         &hdr->saddr, th->source,
1878                                         &hdr->daddr, ntohs(th->dest),
1879                                         inet6_iif(skb), inet6_sdif(skb));
1880         if (sk) {
1881                 skb->sk = sk;
1882                 skb->destructor = sock_edemux;
1883                 if (sk_fullsock(sk)) {
1884                         struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1885
1886                         if (dst)
1887                                 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1888                         if (dst &&
1889                             sk->sk_rx_dst_ifindex == skb->skb_iif)
1890                                 skb_dst_set_noref(skb, dst);
1891                 }
1892         }
1893 }
1894
1895 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1896         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1897         .twsk_unique    = tcp_twsk_unique,
1898         .twsk_destructor = tcp_twsk_destructor,
1899 };
1900
1901 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1902 {
1903         struct ipv6_pinfo *np = inet6_sk(sk);
1904
1905         __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
1906 }
1907
1908 const struct inet_connection_sock_af_ops ipv6_specific = {
1909         .queue_xmit        = inet6_csk_xmit,
1910         .send_check        = tcp_v6_send_check,
1911         .rebuild_header    = inet6_sk_rebuild_header,
1912         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1913         .conn_request      = tcp_v6_conn_request,
1914         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1915         .net_header_len    = sizeof(struct ipv6hdr),
1916         .net_frag_header_len = sizeof(struct frag_hdr),
1917         .setsockopt        = ipv6_setsockopt,
1918         .getsockopt        = ipv6_getsockopt,
1919         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1920         .sockaddr_len      = sizeof(struct sockaddr_in6),
1921         .mtu_reduced       = tcp_v6_mtu_reduced,
1922 };
1923
1924 #ifdef CONFIG_TCP_MD5SIG
1925 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1926         .md5_lookup     =       tcp_v6_md5_lookup,
1927         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1928         .md5_parse      =       tcp_v6_parse_md5_keys,
1929 };
1930 #endif
1931
1932 /*
1933  *      TCP over IPv4 via INET6 API
1934  */
1935 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1936         .queue_xmit        = ip_queue_xmit,
1937         .send_check        = tcp_v4_send_check,
1938         .rebuild_header    = inet_sk_rebuild_header,
1939         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1940         .conn_request      = tcp_v6_conn_request,
1941         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1942         .net_header_len    = sizeof(struct iphdr),
1943         .setsockopt        = ipv6_setsockopt,
1944         .getsockopt        = ipv6_getsockopt,
1945         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1946         .sockaddr_len      = sizeof(struct sockaddr_in6),
1947         .mtu_reduced       = tcp_v4_mtu_reduced,
1948 };
1949
1950 #ifdef CONFIG_TCP_MD5SIG
1951 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1952         .md5_lookup     =       tcp_v4_md5_lookup,
1953         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1954         .md5_parse      =       tcp_v6_parse_md5_keys,
1955 };
1956 #endif
1957
1958 /* NOTE: A lot of things set to zero explicitly by call to
1959  *       sk_alloc() so need not be done here.
1960  */
1961 static int tcp_v6_init_sock(struct sock *sk)
1962 {
1963         struct inet_connection_sock *icsk = inet_csk(sk);
1964
1965         tcp_init_sock(sk);
1966
1967         icsk->icsk_af_ops = &ipv6_specific;
1968
1969 #ifdef CONFIG_TCP_MD5SIG
1970         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1971 #endif
1972
1973         return 0;
1974 }
1975
1976 static void tcp_v6_destroy_sock(struct sock *sk)
1977 {
1978         tcp_v4_destroy_sock(sk);
1979         inet6_destroy_sock(sk);
1980 }
1981
1982 #ifdef CONFIG_PROC_FS
1983 /* Proc filesystem TCPv6 sock list dumping. */
1984 static void get_openreq6(struct seq_file *seq,
1985                          const struct request_sock *req, int i)
1986 {
1987         long ttd = req->rsk_timer.expires - jiffies;
1988         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1989         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1990
1991         if (ttd < 0)
1992                 ttd = 0;
1993
1994         seq_printf(seq,
1995                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1996                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1997                    i,
1998                    src->s6_addr32[0], src->s6_addr32[1],
1999                    src->s6_addr32[2], src->s6_addr32[3],
2000                    inet_rsk(req)->ir_num,
2001                    dest->s6_addr32[0], dest->s6_addr32[1],
2002                    dest->s6_addr32[2], dest->s6_addr32[3],
2003                    ntohs(inet_rsk(req)->ir_rmt_port),
2004                    TCP_SYN_RECV,
2005                    0, 0, /* could print option size, but that is af dependent. */
2006                    1,   /* timers active (only the expire timer) */
2007                    jiffies_to_clock_t(ttd),
2008                    req->num_timeout,
2009                    from_kuid_munged(seq_user_ns(seq),
2010                                     sock_i_uid(req->rsk_listener)),
2011                    0,  /* non standard timer */
2012                    0, /* open_requests have no inode */
2013                    0, req);
2014 }
2015
2016 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2017 {
2018         const struct in6_addr *dest, *src;
2019         __u16 destp, srcp;
2020         int timer_active;
2021         unsigned long timer_expires;
2022         const struct inet_sock *inet = inet_sk(sp);
2023         const struct tcp_sock *tp = tcp_sk(sp);
2024         const struct inet_connection_sock *icsk = inet_csk(sp);
2025         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2026         int rx_queue;
2027         int state;
2028
2029         dest  = &sp->sk_v6_daddr;
2030         src   = &sp->sk_v6_rcv_saddr;
2031         destp = ntohs(inet->inet_dport);
2032         srcp  = ntohs(inet->inet_sport);
2033
2034         if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2035             icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2036             icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2037                 timer_active    = 1;
2038                 timer_expires   = icsk->icsk_timeout;
2039         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2040                 timer_active    = 4;
2041                 timer_expires   = icsk->icsk_timeout;
2042         } else if (timer_pending(&sp->sk_timer)) {
2043                 timer_active    = 2;
2044                 timer_expires   = sp->sk_timer.expires;
2045         } else {
2046                 timer_active    = 0;
2047                 timer_expires = jiffies;
2048         }
2049
2050         state = inet_sk_state_load(sp);
2051         if (state == TCP_LISTEN)
2052                 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2053         else
2054                 /* Because we don't lock the socket,
2055                  * we might find a transient negative value.
2056                  */
2057                 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2058                                       READ_ONCE(tp->copied_seq), 0);
2059
2060         seq_printf(seq,
2061                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2062                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2063                    i,
2064                    src->s6_addr32[0], src->s6_addr32[1],
2065                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2066                    dest->s6_addr32[0], dest->s6_addr32[1],
2067                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2068                    state,
2069                    READ_ONCE(tp->write_seq) - tp->snd_una,
2070                    rx_queue,
2071                    timer_active,
2072                    jiffies_delta_to_clock_t(timer_expires - jiffies),
2073                    icsk->icsk_retransmits,
2074                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2075                    icsk->icsk_probes_out,
2076                    sock_i_ino(sp),
2077                    refcount_read(&sp->sk_refcnt), sp,
2078                    jiffies_to_clock_t(icsk->icsk_rto),
2079                    jiffies_to_clock_t(icsk->icsk_ack.ato),
2080                    (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2081                    tcp_snd_cwnd(tp),
2082                    state == TCP_LISTEN ?
2083                         fastopenq->max_qlen :
2084                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2085                    );
2086 }
2087
2088 static void get_timewait6_sock(struct seq_file *seq,
2089                                struct inet_timewait_sock *tw, int i)
2090 {
2091         long delta = tw->tw_timer.expires - jiffies;
2092         const struct in6_addr *dest, *src;
2093         __u16 destp, srcp;
2094
2095         dest = &tw->tw_v6_daddr;
2096         src  = &tw->tw_v6_rcv_saddr;
2097         destp = ntohs(tw->tw_dport);
2098         srcp  = ntohs(tw->tw_sport);
2099
2100         seq_printf(seq,
2101                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2102                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2103                    i,
2104                    src->s6_addr32[0], src->s6_addr32[1],
2105                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2106                    dest->s6_addr32[0], dest->s6_addr32[1],
2107                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2108                    tw->tw_substate, 0, 0,
2109                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2110                    refcount_read(&tw->tw_refcnt), tw);
2111 }
2112
2113 static int tcp6_seq_show(struct seq_file *seq, void *v)
2114 {
2115         struct tcp_iter_state *st;
2116         struct sock *sk = v;
2117
2118         if (v == SEQ_START_TOKEN) {
2119                 seq_puts(seq,
2120                          "  sl  "
2121                          "local_address                         "
2122                          "remote_address                        "
2123                          "st tx_queue rx_queue tr tm->when retrnsmt"
2124                          "   uid  timeout inode\n");
2125                 goto out;
2126         }
2127         st = seq->private;
2128
2129         if (sk->sk_state == TCP_TIME_WAIT)
2130                 get_timewait6_sock(seq, v, st->num);
2131         else if (sk->sk_state == TCP_NEW_SYN_RECV)
2132                 get_openreq6(seq, v, st->num);
2133         else
2134                 get_tcp6_sock(seq, v, st->num);
2135 out:
2136         return 0;
2137 }
2138
2139 static const struct seq_operations tcp6_seq_ops = {
2140         .show           = tcp6_seq_show,
2141         .start          = tcp_seq_start,
2142         .next           = tcp_seq_next,
2143         .stop           = tcp_seq_stop,
2144 };
2145
2146 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2147         .family         = AF_INET6,
2148 };
2149
2150 int __net_init tcp6_proc_init(struct net *net)
2151 {
2152         if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2153                         sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2154                 return -ENOMEM;
2155         return 0;
2156 }
2157
2158 void tcp6_proc_exit(struct net *net)
2159 {
2160         remove_proc_entry("tcp6", net->proc_net);
2161 }
2162 #endif
2163
2164 struct proto tcpv6_prot = {
2165         .name                   = "TCPv6",
2166         .owner                  = THIS_MODULE,
2167         .close                  = tcp_close,
2168         .pre_connect            = tcp_v6_pre_connect,
2169         .connect                = tcp_v6_connect,
2170         .disconnect             = tcp_disconnect,
2171         .accept                 = inet_csk_accept,
2172         .ioctl                  = tcp_ioctl,
2173         .init                   = tcp_v6_init_sock,
2174         .destroy                = tcp_v6_destroy_sock,
2175         .shutdown               = tcp_shutdown,
2176         .setsockopt             = tcp_setsockopt,
2177         .getsockopt             = tcp_getsockopt,
2178         .bpf_bypass_getsockopt  = tcp_bpf_bypass_getsockopt,
2179         .keepalive              = tcp_set_keepalive,
2180         .recvmsg                = tcp_recvmsg,
2181         .sendmsg                = tcp_sendmsg,
2182         .sendpage               = tcp_sendpage,
2183         .backlog_rcv            = tcp_v6_do_rcv,
2184         .release_cb             = tcp_release_cb,
2185         .hash                   = inet6_hash,
2186         .unhash                 = inet_unhash,
2187         .get_port               = inet_csk_get_port,
2188 #ifdef CONFIG_BPF_SYSCALL
2189         .psock_update_sk_prot   = tcp_bpf_update_proto,
2190 #endif
2191         .enter_memory_pressure  = tcp_enter_memory_pressure,
2192         .leave_memory_pressure  = tcp_leave_memory_pressure,
2193         .stream_memory_free     = tcp_stream_memory_free,
2194         .sockets_allocated      = &tcp_sockets_allocated,
2195         .memory_allocated       = &tcp_memory_allocated,
2196         .memory_pressure        = &tcp_memory_pressure,
2197         .orphan_count           = &tcp_orphan_count,
2198         .sysctl_mem             = sysctl_tcp_mem,
2199         .sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2200         .sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2201         .max_header             = MAX_TCP_HEADER,
2202         .obj_size               = sizeof(struct tcp6_sock),
2203         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
2204         .twsk_prot              = &tcp6_timewait_sock_ops,
2205         .rsk_prot               = &tcp6_request_sock_ops,
2206         .h.hashinfo             = &tcp_hashinfo,
2207         .no_autobind            = true,
2208         .diag_destroy           = tcp_abort,
2209 };
2210 EXPORT_SYMBOL_GPL(tcpv6_prot);
2211
2212 static const struct inet6_protocol tcpv6_protocol = {
2213         .handler        =       tcp_v6_rcv,
2214         .err_handler    =       tcp_v6_err,
2215         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2216 };
2217
2218 static struct inet_protosw tcpv6_protosw = {
2219         .type           =       SOCK_STREAM,
2220         .protocol       =       IPPROTO_TCP,
2221         .prot           =       &tcpv6_prot,
2222         .ops            =       &inet6_stream_ops,
2223         .flags          =       INET_PROTOSW_PERMANENT |
2224                                 INET_PROTOSW_ICSK,
2225 };
2226
2227 static int __net_init tcpv6_net_init(struct net *net)
2228 {
2229         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2230                                     SOCK_RAW, IPPROTO_TCP, net);
2231 }
2232
2233 static void __net_exit tcpv6_net_exit(struct net *net)
2234 {
2235         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2236 }
2237
2238 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2239 {
2240         inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2241 }
2242
2243 static struct pernet_operations tcpv6_net_ops = {
2244         .init       = tcpv6_net_init,
2245         .exit       = tcpv6_net_exit,
2246         .exit_batch = tcpv6_net_exit_batch,
2247 };
2248
2249 int __init tcpv6_init(void)
2250 {
2251         int ret;
2252
2253         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2254         if (ret)
2255                 goto out;
2256
2257         /* register inet6 protocol */
2258         ret = inet6_register_protosw(&tcpv6_protosw);
2259         if (ret)
2260                 goto out_tcpv6_protocol;
2261
2262         ret = register_pernet_subsys(&tcpv6_net_ops);
2263         if (ret)
2264                 goto out_tcpv6_protosw;
2265
2266         ret = mptcpv6_init();
2267         if (ret)
2268                 goto out_tcpv6_pernet_subsys;
2269
2270 out:
2271         return ret;
2272
2273 out_tcpv6_pernet_subsys:
2274         unregister_pernet_subsys(&tcpv6_net_ops);
2275 out_tcpv6_protosw:
2276         inet6_unregister_protosw(&tcpv6_protosw);
2277 out_tcpv6_protocol:
2278         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2279         goto out;
2280 }
2281
2282 void tcpv6_exit(void)
2283 {
2284         unregister_pernet_subsys(&tcpv6_net_ops);
2285         inet6_unregister_protosw(&tcpv6_protosw);
2286         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2287 }