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 INDIRECT_CALLABLE_SCOPE 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_connection_sock *icsk = inet_csk(sk); 150 struct in6_addr *saddr = NULL, *final_p, final; 151 struct inet_timewait_death_row *tcp_death_row; 152 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 153 struct inet_sock *inet = inet_sk(sk); 154 struct tcp_sock *tp = tcp_sk(sk); 155 struct net *net = sock_net(sk); 156 struct ipv6_txoptions *opt; 157 struct dst_entry *dst; 158 struct flowi6 fl6; 159 int addr_type; 160 int err; 161 162 if (addr_len < SIN6_LEN_RFC2133) 163 return -EINVAL; 164 165 if (usin->sin6_family != AF_INET6) 166 return -EAFNOSUPPORT; 167 168 memset(&fl6, 0, sizeof(fl6)); 169 170 if (np->sndflow) { 171 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; 172 IP6_ECN_flow_init(fl6.flowlabel); 173 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 174 struct ip6_flowlabel *flowlabel; 175 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 176 if (IS_ERR(flowlabel)) 177 return -EINVAL; 178 fl6_sock_release(flowlabel); 179 } 180 } 181 182 /* 183 * connect() to INADDR_ANY means loopback (BSD'ism). 184 */ 185 186 if (ipv6_addr_any(&usin->sin6_addr)) { 187 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 188 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 189 &usin->sin6_addr); 190 else 191 usin->sin6_addr = in6addr_loopback; 192 } 193 194 addr_type = ipv6_addr_type(&usin->sin6_addr); 195 196 if (addr_type & IPV6_ADDR_MULTICAST) 197 return -ENETUNREACH; 198 199 if (addr_type&IPV6_ADDR_LINKLOCAL) { 200 if (addr_len >= sizeof(struct sockaddr_in6) && 201 usin->sin6_scope_id) { 202 /* If interface is set while binding, indices 203 * must coincide. 204 */ 205 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id)) 206 return -EINVAL; 207 208 sk->sk_bound_dev_if = usin->sin6_scope_id; 209 } 210 211 /* Connect to link-local address requires an interface */ 212 if (!sk->sk_bound_dev_if) 213 return -EINVAL; 214 } 215 216 if (tp->rx_opt.ts_recent_stamp && 217 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) { 218 tp->rx_opt.ts_recent = 0; 219 tp->rx_opt.ts_recent_stamp = 0; 220 WRITE_ONCE(tp->write_seq, 0); 221 } 222 223 sk->sk_v6_daddr = usin->sin6_addr; 224 np->flow_label = fl6.flowlabel; 225 226 /* 227 * TCP over IPv4 228 */ 229 230 if (addr_type & IPV6_ADDR_MAPPED) { 231 u32 exthdrlen = icsk->icsk_ext_hdr_len; 232 struct sockaddr_in sin; 233 234 if (ipv6_only_sock(sk)) 235 return -ENETUNREACH; 236 237 sin.sin_family = AF_INET; 238 sin.sin_port = usin->sin6_port; 239 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 240 241 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ 242 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped); 243 if (sk_is_mptcp(sk)) 244 mptcpv6_handle_mapped(sk, true); 245 sk->sk_backlog_rcv = tcp_v4_do_rcv; 246 #ifdef CONFIG_TCP_MD5SIG 247 tp->af_specific = &tcp_sock_ipv6_mapped_specific; 248 #endif 249 250 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 251 252 if (err) { 253 icsk->icsk_ext_hdr_len = exthdrlen; 254 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ 255 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific); 256 if (sk_is_mptcp(sk)) 257 mptcpv6_handle_mapped(sk, false); 258 sk->sk_backlog_rcv = tcp_v6_do_rcv; 259 #ifdef CONFIG_TCP_MD5SIG 260 tp->af_specific = &tcp_sock_ipv6_specific; 261 #endif 262 goto failure; 263 } 264 np->saddr = sk->sk_v6_rcv_saddr; 265 266 return err; 267 } 268 269 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 270 saddr = &sk->sk_v6_rcv_saddr; 271 272 fl6.flowi6_proto = IPPROTO_TCP; 273 fl6.daddr = sk->sk_v6_daddr; 274 fl6.saddr = saddr ? *saddr : np->saddr; 275 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label); 276 fl6.flowi6_oif = sk->sk_bound_dev_if; 277 fl6.flowi6_mark = sk->sk_mark; 278 fl6.fl6_dport = usin->sin6_port; 279 fl6.fl6_sport = inet->inet_sport; 280 fl6.flowi6_uid = sk->sk_uid; 281 282 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk)); 283 final_p = fl6_update_dst(&fl6, opt, &final); 284 285 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 286 287 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p); 288 if (IS_ERR(dst)) { 289 err = PTR_ERR(dst); 290 goto failure; 291 } 292 293 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; 294 295 if (!saddr) { 296 saddr = &fl6.saddr; 297 298 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6); 299 if (err) 300 goto failure; 301 } 302 303 /* set the source address */ 304 np->saddr = *saddr; 305 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 306 307 sk->sk_gso_type = SKB_GSO_TCPV6; 308 ip6_dst_store(sk, dst, NULL, NULL); 309 310 icsk->icsk_ext_hdr_len = 0; 311 if (opt) 312 icsk->icsk_ext_hdr_len = opt->opt_flen + 313 opt->opt_nflen; 314 315 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 316 317 inet->inet_dport = usin->sin6_port; 318 319 tcp_set_state(sk, TCP_SYN_SENT); 320 err = inet6_hash_connect(tcp_death_row, sk); 321 if (err) 322 goto late_failure; 323 324 sk_set_txhash(sk); 325 326 if (likely(!tp->repair)) { 327 if (!tp->write_seq) 328 WRITE_ONCE(tp->write_seq, 329 secure_tcpv6_seq(np->saddr.s6_addr32, 330 sk->sk_v6_daddr.s6_addr32, 331 inet->inet_sport, 332 inet->inet_dport)); 333 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32, 334 sk->sk_v6_daddr.s6_addr32); 335 } 336 337 if (tcp_fastopen_defer_connect(sk, &err)) 338 return err; 339 if (err) 340 goto late_failure; 341 342 err = tcp_connect(sk); 343 if (err) 344 goto late_failure; 345 346 return 0; 347 348 late_failure: 349 tcp_set_state(sk, TCP_CLOSE); 350 inet_bhash2_reset_saddr(sk); 351 failure: 352 inet->inet_dport = 0; 353 sk->sk_route_caps = 0; 354 return err; 355 } 356 357 static void tcp_v6_mtu_reduced(struct sock *sk) 358 { 359 struct dst_entry *dst; 360 u32 mtu; 361 362 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) 363 return; 364 365 mtu = READ_ONCE(tcp_sk(sk)->mtu_info); 366 367 /* Drop requests trying to increase our current mss. 368 * Check done in __ip6_rt_update_pmtu() is too late. 369 */ 370 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache) 371 return; 372 373 dst = inet6_csk_update_pmtu(sk, mtu); 374 if (!dst) 375 return; 376 377 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 378 tcp_sync_mss(sk, dst_mtu(dst)); 379 tcp_simple_retransmit(sk); 380 } 381 } 382 383 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 384 u8 type, u8 code, int offset, __be32 info) 385 { 386 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 387 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); 388 struct net *net = dev_net(skb->dev); 389 struct request_sock *fastopen; 390 struct ipv6_pinfo *np; 391 struct tcp_sock *tp; 392 __u32 seq, snd_una; 393 struct sock *sk; 394 bool fatal; 395 int err; 396 397 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 398 &hdr->daddr, th->dest, 399 &hdr->saddr, ntohs(th->source), 400 skb->dev->ifindex, inet6_sdif(skb)); 401 402 if (!sk) { 403 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 404 ICMP6_MIB_INERRORS); 405 return -ENOENT; 406 } 407 408 if (sk->sk_state == TCP_TIME_WAIT) { 409 inet_twsk_put(inet_twsk(sk)); 410 return 0; 411 } 412 seq = ntohl(th->seq); 413 fatal = icmpv6_err_convert(type, code, &err); 414 if (sk->sk_state == TCP_NEW_SYN_RECV) { 415 tcp_req_err(sk, seq, fatal); 416 return 0; 417 } 418 419 bh_lock_sock(sk); 420 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) 421 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 422 423 if (sk->sk_state == TCP_CLOSE) 424 goto out; 425 426 if (static_branch_unlikely(&ip6_min_hopcount)) { 427 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ 428 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) { 429 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 430 goto out; 431 } 432 } 433 434 tp = tcp_sk(sk); 435 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ 436 fastopen = rcu_dereference(tp->fastopen_rsk); 437 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; 438 if (sk->sk_state != TCP_LISTEN && 439 !between(seq, snd_una, tp->snd_nxt)) { 440 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 441 goto out; 442 } 443 444 np = tcp_inet6_sk(sk); 445 446 if (type == NDISC_REDIRECT) { 447 if (!sock_owned_by_user(sk)) { 448 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie); 449 450 if (dst) 451 dst->ops->redirect(dst, sk, skb); 452 } 453 goto out; 454 } 455 456 if (type == ICMPV6_PKT_TOOBIG) { 457 u32 mtu = ntohl(info); 458 459 /* We are not interested in TCP_LISTEN and open_requests 460 * (SYN-ACKs send out by Linux are always <576bytes so 461 * they should go through unfragmented). 462 */ 463 if (sk->sk_state == TCP_LISTEN) 464 goto out; 465 466 if (!ip6_sk_accept_pmtu(sk)) 467 goto out; 468 469 if (mtu < IPV6_MIN_MTU) 470 goto out; 471 472 WRITE_ONCE(tp->mtu_info, mtu); 473 474 if (!sock_owned_by_user(sk)) 475 tcp_v6_mtu_reduced(sk); 476 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, 477 &sk->sk_tsq_flags)) 478 sock_hold(sk); 479 goto out; 480 } 481 482 483 /* Might be for an request_sock */ 484 switch (sk->sk_state) { 485 case TCP_SYN_SENT: 486 case TCP_SYN_RECV: 487 /* Only in fast or simultaneous open. If a fast open socket is 488 * already accepted it is treated as a connected one below. 489 */ 490 if (fastopen && !fastopen->sk) 491 break; 492 493 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th); 494 495 if (!sock_owned_by_user(sk)) { 496 WRITE_ONCE(sk->sk_err, err); 497 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ 498 499 tcp_done(sk); 500 } else { 501 WRITE_ONCE(sk->sk_err_soft, err); 502 } 503 goto out; 504 case TCP_LISTEN: 505 break; 506 default: 507 /* check if this ICMP message allows revert of backoff. 508 * (see RFC 6069) 509 */ 510 if (!fastopen && type == ICMPV6_DEST_UNREACH && 511 code == ICMPV6_NOROUTE) 512 tcp_ld_RTO_revert(sk, seq); 513 } 514 515 if (!sock_owned_by_user(sk) && np->recverr) { 516 WRITE_ONCE(sk->sk_err, err); 517 sk_error_report(sk); 518 } else { 519 WRITE_ONCE(sk->sk_err_soft, err); 520 } 521 out: 522 bh_unlock_sock(sk); 523 sock_put(sk); 524 return 0; 525 } 526 527 528 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, 529 struct flowi *fl, 530 struct request_sock *req, 531 struct tcp_fastopen_cookie *foc, 532 enum tcp_synack_type synack_type, 533 struct sk_buff *syn_skb) 534 { 535 struct inet_request_sock *ireq = inet_rsk(req); 536 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 537 struct ipv6_txoptions *opt; 538 struct flowi6 *fl6 = &fl->u.ip6; 539 struct sk_buff *skb; 540 int err = -ENOMEM; 541 u8 tclass; 542 543 /* First, grab a route. */ 544 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req, 545 IPPROTO_TCP)) == NULL) 546 goto done; 547 548 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb); 549 550 if (skb) { 551 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, 552 &ireq->ir_v6_rmt_addr); 553 554 fl6->daddr = ireq->ir_v6_rmt_addr; 555 if (np->repflow && ireq->pktopts) 556 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); 557 558 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ? 559 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) | 560 (np->tclass & INET_ECN_MASK) : 561 np->tclass; 562 563 if (!INET_ECN_is_capable(tclass) && 564 tcp_bpf_ca_needs_ecn((struct sock *)req)) 565 tclass |= INET_ECN_ECT_0; 566 567 rcu_read_lock(); 568 opt = ireq->ipv6_opt; 569 if (!opt) 570 opt = rcu_dereference(np->opt); 571 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt, 572 tclass, sk->sk_priority); 573 rcu_read_unlock(); 574 err = net_xmit_eval(err); 575 } 576 577 done: 578 return err; 579 } 580 581 582 static void tcp_v6_reqsk_destructor(struct request_sock *req) 583 { 584 kfree(inet_rsk(req)->ipv6_opt); 585 consume_skb(inet_rsk(req)->pktopts); 586 } 587 588 #ifdef CONFIG_TCP_MD5SIG 589 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, 590 const struct in6_addr *addr, 591 int l3index) 592 { 593 return tcp_md5_do_lookup(sk, l3index, 594 (union tcp_md5_addr *)addr, AF_INET6); 595 } 596 597 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk, 598 const struct sock *addr_sk) 599 { 600 int l3index; 601 602 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 603 addr_sk->sk_bound_dev_if); 604 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr, 605 l3index); 606 } 607 608 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname, 609 sockptr_t optval, int optlen) 610 { 611 struct tcp_md5sig cmd; 612 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; 613 int l3index = 0; 614 u8 prefixlen; 615 u8 flags; 616 617 if (optlen < sizeof(cmd)) 618 return -EINVAL; 619 620 if (copy_from_sockptr(&cmd, optval, sizeof(cmd))) 621 return -EFAULT; 622 623 if (sin6->sin6_family != AF_INET6) 624 return -EINVAL; 625 626 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX; 627 628 if (optname == TCP_MD5SIG_EXT && 629 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { 630 prefixlen = cmd.tcpm_prefixlen; 631 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) && 632 prefixlen > 32)) 633 return -EINVAL; 634 } else { 635 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128; 636 } 637 638 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex && 639 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) { 640 struct net_device *dev; 641 642 rcu_read_lock(); 643 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex); 644 if (dev && netif_is_l3_master(dev)) 645 l3index = dev->ifindex; 646 rcu_read_unlock(); 647 648 /* ok to reference set/not set outside of rcu; 649 * right now device MUST be an L3 master 650 */ 651 if (!dev || !l3index) 652 return -EINVAL; 653 } 654 655 if (!cmd.tcpm_keylen) { 656 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 657 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 658 AF_INET, prefixlen, 659 l3index, flags); 660 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 661 AF_INET6, prefixlen, l3index, flags); 662 } 663 664 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 665 return -EINVAL; 666 667 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 668 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 669 AF_INET, prefixlen, l3index, flags, 670 cmd.tcpm_key, cmd.tcpm_keylen); 671 672 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 673 AF_INET6, prefixlen, l3index, flags, 674 cmd.tcpm_key, cmd.tcpm_keylen); 675 } 676 677 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp, 678 const struct in6_addr *daddr, 679 const struct in6_addr *saddr, 680 const struct tcphdr *th, int nbytes) 681 { 682 struct tcp6_pseudohdr *bp; 683 struct scatterlist sg; 684 struct tcphdr *_th; 685 686 bp = hp->scratch; 687 /* 1. TCP pseudo-header (RFC2460) */ 688 bp->saddr = *saddr; 689 bp->daddr = *daddr; 690 bp->protocol = cpu_to_be32(IPPROTO_TCP); 691 bp->len = cpu_to_be32(nbytes); 692 693 _th = (struct tcphdr *)(bp + 1); 694 memcpy(_th, th, sizeof(*th)); 695 _th->check = 0; 696 697 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); 698 ahash_request_set_crypt(hp->md5_req, &sg, NULL, 699 sizeof(*bp) + sizeof(*th)); 700 return crypto_ahash_update(hp->md5_req); 701 } 702 703 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 704 const struct in6_addr *daddr, struct in6_addr *saddr, 705 const struct tcphdr *th) 706 { 707 struct tcp_md5sig_pool *hp; 708 struct ahash_request *req; 709 710 hp = tcp_get_md5sig_pool(); 711 if (!hp) 712 goto clear_hash_noput; 713 req = hp->md5_req; 714 715 if (crypto_ahash_init(req)) 716 goto clear_hash; 717 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) 718 goto clear_hash; 719 if (tcp_md5_hash_key(hp, key)) 720 goto clear_hash; 721 ahash_request_set_crypt(req, NULL, md5_hash, 0); 722 if (crypto_ahash_final(req)) 723 goto clear_hash; 724 725 tcp_put_md5sig_pool(); 726 return 0; 727 728 clear_hash: 729 tcp_put_md5sig_pool(); 730 clear_hash_noput: 731 memset(md5_hash, 0, 16); 732 return 1; 733 } 734 735 static int tcp_v6_md5_hash_skb(char *md5_hash, 736 const struct tcp_md5sig_key *key, 737 const struct sock *sk, 738 const struct sk_buff *skb) 739 { 740 const struct in6_addr *saddr, *daddr; 741 struct tcp_md5sig_pool *hp; 742 struct ahash_request *req; 743 const struct tcphdr *th = tcp_hdr(skb); 744 745 if (sk) { /* valid for establish/request sockets */ 746 saddr = &sk->sk_v6_rcv_saddr; 747 daddr = &sk->sk_v6_daddr; 748 } else { 749 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 750 saddr = &ip6h->saddr; 751 daddr = &ip6h->daddr; 752 } 753 754 hp = tcp_get_md5sig_pool(); 755 if (!hp) 756 goto clear_hash_noput; 757 req = hp->md5_req; 758 759 if (crypto_ahash_init(req)) 760 goto clear_hash; 761 762 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len)) 763 goto clear_hash; 764 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 765 goto clear_hash; 766 if (tcp_md5_hash_key(hp, key)) 767 goto clear_hash; 768 ahash_request_set_crypt(req, NULL, md5_hash, 0); 769 if (crypto_ahash_final(req)) 770 goto clear_hash; 771 772 tcp_put_md5sig_pool(); 773 return 0; 774 775 clear_hash: 776 tcp_put_md5sig_pool(); 777 clear_hash_noput: 778 memset(md5_hash, 0, 16); 779 return 1; 780 } 781 782 #endif 783 784 static void tcp_v6_init_req(struct request_sock *req, 785 const struct sock *sk_listener, 786 struct sk_buff *skb) 787 { 788 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 789 struct inet_request_sock *ireq = inet_rsk(req); 790 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener); 791 792 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; 793 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; 794 795 /* So that link locals have meaning */ 796 if ((!sk_listener->sk_bound_dev_if || l3_slave) && 797 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) 798 ireq->ir_iif = tcp_v6_iif(skb); 799 800 if (!TCP_SKB_CB(skb)->tcp_tw_isn && 801 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) || 802 np->rxopt.bits.rxinfo || 803 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || 804 np->rxopt.bits.rxohlim || np->repflow)) { 805 refcount_inc(&skb->users); 806 ireq->pktopts = skb; 807 } 808 } 809 810 static struct dst_entry *tcp_v6_route_req(const struct sock *sk, 811 struct sk_buff *skb, 812 struct flowi *fl, 813 struct request_sock *req) 814 { 815 tcp_v6_init_req(req, sk, skb); 816 817 if (security_inet_conn_request(sk, skb, req)) 818 return NULL; 819 820 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP); 821 } 822 823 struct request_sock_ops tcp6_request_sock_ops __read_mostly = { 824 .family = AF_INET6, 825 .obj_size = sizeof(struct tcp6_request_sock), 826 .rtx_syn_ack = tcp_rtx_synack, 827 .send_ack = tcp_v6_reqsk_send_ack, 828 .destructor = tcp_v6_reqsk_destructor, 829 .send_reset = tcp_v6_send_reset, 830 .syn_ack_timeout = tcp_syn_ack_timeout, 831 }; 832 833 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { 834 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 835 sizeof(struct ipv6hdr), 836 #ifdef CONFIG_TCP_MD5SIG 837 .req_md5_lookup = tcp_v6_md5_lookup, 838 .calc_md5_hash = tcp_v6_md5_hash_skb, 839 #endif 840 #ifdef CONFIG_SYN_COOKIES 841 .cookie_init_seq = cookie_v6_init_sequence, 842 #endif 843 .route_req = tcp_v6_route_req, 844 .init_seq = tcp_v6_init_seq, 845 .init_ts_off = tcp_v6_init_ts_off, 846 .send_synack = tcp_v6_send_synack, 847 }; 848 849 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, 850 u32 ack, u32 win, u32 tsval, u32 tsecr, 851 int oif, struct tcp_md5sig_key *key, int rst, 852 u8 tclass, __be32 label, u32 priority, u32 txhash) 853 { 854 const struct tcphdr *th = tcp_hdr(skb); 855 struct tcphdr *t1; 856 struct sk_buff *buff; 857 struct flowi6 fl6; 858 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 859 struct sock *ctl_sk = net->ipv6.tcp_sk; 860 unsigned int tot_len = sizeof(struct tcphdr); 861 __be32 mrst = 0, *topt; 862 struct dst_entry *dst; 863 __u32 mark = 0; 864 865 if (tsecr) 866 tot_len += TCPOLEN_TSTAMP_ALIGNED; 867 #ifdef CONFIG_TCP_MD5SIG 868 if (key) 869 tot_len += TCPOLEN_MD5SIG_ALIGNED; 870 #endif 871 872 #ifdef CONFIG_MPTCP 873 if (rst && !key) { 874 mrst = mptcp_reset_option(skb); 875 876 if (mrst) 877 tot_len += sizeof(__be32); 878 } 879 #endif 880 881 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC); 882 if (!buff) 883 return; 884 885 skb_reserve(buff, MAX_TCP_HEADER); 886 887 t1 = skb_push(buff, tot_len); 888 skb_reset_transport_header(buff); 889 890 /* Swap the send and the receive. */ 891 memset(t1, 0, sizeof(*t1)); 892 t1->dest = th->source; 893 t1->source = th->dest; 894 t1->doff = tot_len / 4; 895 t1->seq = htonl(seq); 896 t1->ack_seq = htonl(ack); 897 t1->ack = !rst || !th->ack; 898 t1->rst = rst; 899 t1->window = htons(win); 900 901 topt = (__be32 *)(t1 + 1); 902 903 if (tsecr) { 904 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 905 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 906 *topt++ = htonl(tsval); 907 *topt++ = htonl(tsecr); 908 } 909 910 if (mrst) 911 *topt++ = mrst; 912 913 #ifdef CONFIG_TCP_MD5SIG 914 if (key) { 915 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 916 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); 917 tcp_v6_md5_hash_hdr((__u8 *)topt, key, 918 &ipv6_hdr(skb)->saddr, 919 &ipv6_hdr(skb)->daddr, t1); 920 } 921 #endif 922 923 memset(&fl6, 0, sizeof(fl6)); 924 fl6.daddr = ipv6_hdr(skb)->saddr; 925 fl6.saddr = ipv6_hdr(skb)->daddr; 926 fl6.flowlabel = label; 927 928 buff->ip_summed = CHECKSUM_PARTIAL; 929 930 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr); 931 932 fl6.flowi6_proto = IPPROTO_TCP; 933 if (rt6_need_strict(&fl6.daddr) && !oif) 934 fl6.flowi6_oif = tcp_v6_iif(skb); 935 else { 936 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 937 oif = skb->skb_iif; 938 939 fl6.flowi6_oif = oif; 940 } 941 942 if (sk) { 943 if (sk->sk_state == TCP_TIME_WAIT) 944 mark = inet_twsk(sk)->tw_mark; 945 else 946 mark = sk->sk_mark; 947 skb_set_delivery_time(buff, tcp_transmit_time(sk), true); 948 } 949 if (txhash) { 950 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */ 951 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4); 952 } 953 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark; 954 fl6.fl6_dport = t1->dest; 955 fl6.fl6_sport = t1->source; 956 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); 957 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6)); 958 959 /* Pass a socket to ip6_dst_lookup either it is for RST 960 * Underlying function will use this to retrieve the network 961 * namespace 962 */ 963 if (sk && sk->sk_state != TCP_TIME_WAIT) 964 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/ 965 else 966 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL); 967 if (!IS_ERR(dst)) { 968 skb_dst_set(buff, dst); 969 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, 970 tclass & ~INET_ECN_MASK, priority); 971 TCP_INC_STATS(net, TCP_MIB_OUTSEGS); 972 if (rst) 973 TCP_INC_STATS(net, TCP_MIB_OUTRSTS); 974 return; 975 } 976 977 kfree_skb(buff); 978 } 979 980 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) 981 { 982 const struct tcphdr *th = tcp_hdr(skb); 983 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 984 u32 seq = 0, ack_seq = 0; 985 struct tcp_md5sig_key *key = NULL; 986 #ifdef CONFIG_TCP_MD5SIG 987 const __u8 *hash_location = NULL; 988 unsigned char newhash[16]; 989 int genhash; 990 struct sock *sk1 = NULL; 991 #endif 992 __be32 label = 0; 993 u32 priority = 0; 994 struct net *net; 995 u32 txhash = 0; 996 int oif = 0; 997 998 if (th->rst) 999 return; 1000 1001 /* If sk not NULL, it means we did a successful lookup and incoming 1002 * route had to be correct. prequeue might have dropped our dst. 1003 */ 1004 if (!sk && !ipv6_unicast_destination(skb)) 1005 return; 1006 1007 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 1008 #ifdef CONFIG_TCP_MD5SIG 1009 rcu_read_lock(); 1010 hash_location = tcp_parse_md5sig_option(th); 1011 if (sk && sk_fullsock(sk)) { 1012 int l3index; 1013 1014 /* sdif set, means packet ingressed via a device 1015 * in an L3 domain and inet_iif is set to it. 1016 */ 1017 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1018 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index); 1019 } else if (hash_location) { 1020 int dif = tcp_v6_iif_l3_slave(skb); 1021 int sdif = tcp_v6_sdif(skb); 1022 int l3index; 1023 1024 /* 1025 * active side is lost. Try to find listening socket through 1026 * source port, and then find md5 key through listening socket. 1027 * we are not loose security here: 1028 * Incoming packet is checked with md5 hash with finding key, 1029 * no RST generated if md5 hash doesn't match. 1030 */ 1031 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1032 NULL, 0, &ipv6h->saddr, th->source, 1033 &ipv6h->daddr, ntohs(th->source), 1034 dif, sdif); 1035 if (!sk1) 1036 goto out; 1037 1038 /* sdif set, means packet ingressed via a device 1039 * in an L3 domain and dif is set to it. 1040 */ 1041 l3index = tcp_v6_sdif(skb) ? dif : 0; 1042 1043 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index); 1044 if (!key) 1045 goto out; 1046 1047 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb); 1048 if (genhash || memcmp(hash_location, newhash, 16) != 0) 1049 goto out; 1050 } 1051 #endif 1052 1053 if (th->ack) 1054 seq = ntohl(th->ack_seq); 1055 else 1056 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - 1057 (th->doff << 2); 1058 1059 if (sk) { 1060 oif = sk->sk_bound_dev_if; 1061 if (sk_fullsock(sk)) { 1062 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1063 1064 trace_tcp_send_reset(sk, skb); 1065 if (np->repflow) 1066 label = ip6_flowlabel(ipv6h); 1067 priority = sk->sk_priority; 1068 txhash = sk->sk_hash; 1069 } 1070 if (sk->sk_state == TCP_TIME_WAIT) { 1071 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel); 1072 priority = inet_twsk(sk)->tw_priority; 1073 txhash = inet_twsk(sk)->tw_txhash; 1074 } 1075 } else { 1076 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET) 1077 label = ip6_flowlabel(ipv6h); 1078 } 1079 1080 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 1081 ipv6_get_dsfield(ipv6h), label, priority, txhash); 1082 1083 #ifdef CONFIG_TCP_MD5SIG 1084 out: 1085 rcu_read_unlock(); 1086 #endif 1087 } 1088 1089 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, 1090 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, 1091 struct tcp_md5sig_key *key, u8 tclass, 1092 __be32 label, u32 priority, u32 txhash) 1093 { 1094 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0, 1095 tclass, label, priority, txhash); 1096 } 1097 1098 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) 1099 { 1100 struct inet_timewait_sock *tw = inet_twsk(sk); 1101 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 1102 1103 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 1104 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 1105 tcp_time_stamp_raw() + tcptw->tw_ts_offset, 1106 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw), 1107 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority, 1108 tw->tw_txhash); 1109 1110 inet_twsk_put(tw); 1111 } 1112 1113 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 1114 struct request_sock *req) 1115 { 1116 int l3index; 1117 1118 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1119 1120 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV 1121 * sk->sk_state == TCP_SYN_RECV -> for Fast Open. 1122 */ 1123 /* RFC 7323 2.3 1124 * The window field (SEG.WND) of every outgoing segment, with the 1125 * exception of <SYN> segments, MUST be right-shifted by 1126 * Rcv.Wind.Shift bits: 1127 */ 1128 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ? 1129 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, 1130 tcp_rsk(req)->rcv_nxt, 1131 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, 1132 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off, 1133 req->ts_recent, sk->sk_bound_dev_if, 1134 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index), 1135 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority, 1136 tcp_rsk(req)->txhash); 1137 } 1138 1139 1140 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) 1141 { 1142 #ifdef CONFIG_SYN_COOKIES 1143 const struct tcphdr *th = tcp_hdr(skb); 1144 1145 if (!th->syn) 1146 sk = cookie_v6_check(sk, skb); 1147 #endif 1148 return sk; 1149 } 1150 1151 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 1152 struct tcphdr *th, u32 *cookie) 1153 { 1154 u16 mss = 0; 1155 #ifdef CONFIG_SYN_COOKIES 1156 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops, 1157 &tcp_request_sock_ipv6_ops, sk, th); 1158 if (mss) { 1159 *cookie = __cookie_v6_init_sequence(iph, th, &mss); 1160 tcp_synq_overflow(sk); 1161 } 1162 #endif 1163 return mss; 1164 } 1165 1166 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 1167 { 1168 if (skb->protocol == htons(ETH_P_IP)) 1169 return tcp_v4_conn_request(sk, skb); 1170 1171 if (!ipv6_unicast_destination(skb)) 1172 goto drop; 1173 1174 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) { 1175 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); 1176 return 0; 1177 } 1178 1179 return tcp_conn_request(&tcp6_request_sock_ops, 1180 &tcp_request_sock_ipv6_ops, sk, skb); 1181 1182 drop: 1183 tcp_listendrop(sk); 1184 return 0; /* don't send reset */ 1185 } 1186 1187 static void tcp_v6_restore_cb(struct sk_buff *skb) 1188 { 1189 /* We need to move header back to the beginning if xfrm6_policy_check() 1190 * and tcp_v6_fill_cb() are going to be called again. 1191 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there. 1192 */ 1193 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, 1194 sizeof(struct inet6_skb_parm)); 1195 } 1196 1197 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 1198 struct request_sock *req, 1199 struct dst_entry *dst, 1200 struct request_sock *req_unhash, 1201 bool *own_req) 1202 { 1203 struct inet_request_sock *ireq; 1204 struct ipv6_pinfo *newnp; 1205 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1206 struct ipv6_txoptions *opt; 1207 struct inet_sock *newinet; 1208 bool found_dup_sk = false; 1209 struct tcp_sock *newtp; 1210 struct sock *newsk; 1211 #ifdef CONFIG_TCP_MD5SIG 1212 struct tcp_md5sig_key *key; 1213 int l3index; 1214 #endif 1215 struct flowi6 fl6; 1216 1217 if (skb->protocol == htons(ETH_P_IP)) { 1218 /* 1219 * v6 mapped 1220 */ 1221 1222 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, 1223 req_unhash, own_req); 1224 1225 if (!newsk) 1226 return NULL; 1227 1228 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1229 1230 newnp = tcp_inet6_sk(newsk); 1231 newtp = tcp_sk(newsk); 1232 1233 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1234 1235 newnp->saddr = newsk->sk_v6_rcv_saddr; 1236 1237 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; 1238 if (sk_is_mptcp(newsk)) 1239 mptcpv6_handle_mapped(newsk, true); 1240 newsk->sk_backlog_rcv = tcp_v4_do_rcv; 1241 #ifdef CONFIG_TCP_MD5SIG 1242 newtp->af_specific = &tcp_sock_ipv6_mapped_specific; 1243 #endif 1244 1245 newnp->ipv6_mc_list = NULL; 1246 newnp->ipv6_ac_list = NULL; 1247 newnp->ipv6_fl_list = NULL; 1248 newnp->pktoptions = NULL; 1249 newnp->opt = NULL; 1250 newnp->mcast_oif = inet_iif(skb); 1251 newnp->mcast_hops = ip_hdr(skb)->ttl; 1252 newnp->rcv_flowinfo = 0; 1253 if (np->repflow) 1254 newnp->flow_label = 0; 1255 1256 /* 1257 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 1258 * here, tcp_create_openreq_child now does this for us, see the comment in 1259 * that function for the gory details. -acme 1260 */ 1261 1262 /* It is tricky place. Until this moment IPv4 tcp 1263 worked with IPv6 icsk.icsk_af_ops. 1264 Sync it now. 1265 */ 1266 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 1267 1268 return newsk; 1269 } 1270 1271 ireq = inet_rsk(req); 1272 1273 if (sk_acceptq_is_full(sk)) 1274 goto out_overflow; 1275 1276 if (!dst) { 1277 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP); 1278 if (!dst) 1279 goto out; 1280 } 1281 1282 newsk = tcp_create_openreq_child(sk, req, skb); 1283 if (!newsk) 1284 goto out_nonewsk; 1285 1286 /* 1287 * No need to charge this sock to the relevant IPv6 refcnt debug socks 1288 * count here, tcp_create_openreq_child now does this for us, see the 1289 * comment in that function for the gory details. -acme 1290 */ 1291 1292 newsk->sk_gso_type = SKB_GSO_TCPV6; 1293 ip6_dst_store(newsk, dst, NULL, NULL); 1294 inet6_sk_rx_dst_set(newsk, skb); 1295 1296 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1297 1298 newtp = tcp_sk(newsk); 1299 newinet = inet_sk(newsk); 1300 newnp = tcp_inet6_sk(newsk); 1301 1302 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1303 1304 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr; 1305 newnp->saddr = ireq->ir_v6_loc_addr; 1306 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr; 1307 newsk->sk_bound_dev_if = ireq->ir_iif; 1308 1309 /* Now IPv6 options... 1310 1311 First: no IPv4 options. 1312 */ 1313 newinet->inet_opt = NULL; 1314 newnp->ipv6_mc_list = NULL; 1315 newnp->ipv6_ac_list = NULL; 1316 newnp->ipv6_fl_list = NULL; 1317 1318 /* Clone RX bits */ 1319 newnp->rxopt.all = np->rxopt.all; 1320 1321 newnp->pktoptions = NULL; 1322 newnp->opt = NULL; 1323 newnp->mcast_oif = tcp_v6_iif(skb); 1324 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1325 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); 1326 if (np->repflow) 1327 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); 1328 1329 /* Set ToS of the new socket based upon the value of incoming SYN. 1330 * ECT bits are set later in tcp_init_transfer(). 1331 */ 1332 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)) 1333 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK; 1334 1335 /* Clone native IPv6 options from listening socket (if any) 1336 1337 Yes, keeping reference count would be much more clever, 1338 but we make one more one thing there: reattach optmem 1339 to newsk. 1340 */ 1341 opt = ireq->ipv6_opt; 1342 if (!opt) 1343 opt = rcu_dereference(np->opt); 1344 if (opt) { 1345 opt = ipv6_dup_options(newsk, opt); 1346 RCU_INIT_POINTER(newnp->opt, opt); 1347 } 1348 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1349 if (opt) 1350 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen + 1351 opt->opt_flen; 1352 1353 tcp_ca_openreq_child(newsk, dst); 1354 1355 tcp_sync_mss(newsk, dst_mtu(dst)); 1356 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst)); 1357 1358 tcp_initialize_rcv_mss(newsk); 1359 1360 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 1361 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 1362 1363 #ifdef CONFIG_TCP_MD5SIG 1364 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif); 1365 1366 /* Copy over the MD5 key from the original socket */ 1367 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index); 1368 if (key) { 1369 const union tcp_md5_addr *addr; 1370 1371 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr; 1372 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) { 1373 inet_csk_prepare_forced_close(newsk); 1374 tcp_done(newsk); 1375 goto out; 1376 } 1377 } 1378 #endif 1379 1380 if (__inet_inherit_port(sk, newsk) < 0) { 1381 inet_csk_prepare_forced_close(newsk); 1382 tcp_done(newsk); 1383 goto out; 1384 } 1385 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), 1386 &found_dup_sk); 1387 if (*own_req) { 1388 tcp_move_syn(newtp, req); 1389 1390 /* Clone pktoptions received with SYN, if we own the req */ 1391 if (ireq->pktopts) { 1392 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk); 1393 consume_skb(ireq->pktopts); 1394 ireq->pktopts = NULL; 1395 if (newnp->pktoptions) 1396 tcp_v6_restore_cb(newnp->pktoptions); 1397 } 1398 } else { 1399 if (!req_unhash && found_dup_sk) { 1400 /* This code path should only be executed in the 1401 * syncookie case only 1402 */ 1403 bh_unlock_sock(newsk); 1404 sock_put(newsk); 1405 newsk = NULL; 1406 } 1407 } 1408 1409 return newsk; 1410 1411 out_overflow: 1412 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1413 out_nonewsk: 1414 dst_release(dst); 1415 out: 1416 tcp_listendrop(sk); 1417 return NULL; 1418 } 1419 1420 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 1421 u32)); 1422 /* The socket must have it's spinlock held when we get 1423 * here, unless it is a TCP_LISTEN socket. 1424 * 1425 * We have a potential double-lock case here, so even when 1426 * doing backlog processing we use the BH locking scheme. 1427 * This is because we cannot sleep with the original spinlock 1428 * held. 1429 */ 1430 INDIRECT_CALLABLE_SCOPE 1431 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 1432 { 1433 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1434 struct sk_buff *opt_skb = NULL; 1435 enum skb_drop_reason reason; 1436 struct tcp_sock *tp; 1437 1438 /* Imagine: socket is IPv6. IPv4 packet arrives, 1439 goes to IPv4 receive handler and backlogged. 1440 From backlog it always goes here. Kerboom... 1441 Fortunately, tcp_rcv_established and rcv_established 1442 handle them correctly, but it is not case with 1443 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK 1444 */ 1445 1446 if (skb->protocol == htons(ETH_P_IP)) 1447 return tcp_v4_do_rcv(sk, skb); 1448 1449 /* 1450 * socket locking is here for SMP purposes as backlog rcv 1451 * is currently called with bh processing disabled. 1452 */ 1453 1454 /* Do Stevens' IPV6_PKTOPTIONS. 1455 1456 Yes, guys, it is the only place in our code, where we 1457 may make it not affecting IPv4. 1458 The rest of code is protocol independent, 1459 and I do not like idea to uglify IPv4. 1460 1461 Actually, all the idea behind IPV6_PKTOPTIONS 1462 looks not very well thought. For now we latch 1463 options, received in the last packet, enqueued 1464 by tcp. Feel free to propose better solution. 1465 --ANK (980728) 1466 */ 1467 if (np->rxopt.all) 1468 opt_skb = skb_clone_and_charge_r(skb, sk); 1469 1470 reason = SKB_DROP_REASON_NOT_SPECIFIED; 1471 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1472 struct dst_entry *dst; 1473 1474 dst = rcu_dereference_protected(sk->sk_rx_dst, 1475 lockdep_sock_is_held(sk)); 1476 1477 sock_rps_save_rxhash(sk, skb); 1478 sk_mark_napi_id(sk, skb); 1479 if (dst) { 1480 if (sk->sk_rx_dst_ifindex != skb->skb_iif || 1481 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check, 1482 dst, sk->sk_rx_dst_cookie) == NULL) { 1483 RCU_INIT_POINTER(sk->sk_rx_dst, NULL); 1484 dst_release(dst); 1485 } 1486 } 1487 1488 tcp_rcv_established(sk, skb); 1489 if (opt_skb) 1490 goto ipv6_pktoptions; 1491 return 0; 1492 } 1493 1494 if (tcp_checksum_complete(skb)) 1495 goto csum_err; 1496 1497 if (sk->sk_state == TCP_LISTEN) { 1498 struct sock *nsk = tcp_v6_cookie_check(sk, skb); 1499 1500 if (!nsk) 1501 goto discard; 1502 1503 if (nsk != sk) { 1504 if (tcp_child_process(sk, nsk, skb)) 1505 goto reset; 1506 if (opt_skb) 1507 __kfree_skb(opt_skb); 1508 return 0; 1509 } 1510 } else 1511 sock_rps_save_rxhash(sk, skb); 1512 1513 if (tcp_rcv_state_process(sk, skb)) 1514 goto reset; 1515 if (opt_skb) 1516 goto ipv6_pktoptions; 1517 return 0; 1518 1519 reset: 1520 tcp_v6_send_reset(sk, skb); 1521 discard: 1522 if (opt_skb) 1523 __kfree_skb(opt_skb); 1524 kfree_skb_reason(skb, reason); 1525 return 0; 1526 csum_err: 1527 reason = SKB_DROP_REASON_TCP_CSUM; 1528 trace_tcp_bad_csum(skb); 1529 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); 1530 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); 1531 goto discard; 1532 1533 1534 ipv6_pktoptions: 1535 /* Do you ask, what is it? 1536 1537 1. skb was enqueued by tcp. 1538 2. skb is added to tail of read queue, rather than out of order. 1539 3. socket is not in passive state. 1540 4. Finally, it really contains options, which user wants to receive. 1541 */ 1542 tp = tcp_sk(sk); 1543 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && 1544 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { 1545 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) 1546 np->mcast_oif = tcp_v6_iif(opt_skb); 1547 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) 1548 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit; 1549 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) 1550 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb)); 1551 if (np->repflow) 1552 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb)); 1553 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) { 1554 tcp_v6_restore_cb(opt_skb); 1555 opt_skb = xchg(&np->pktoptions, opt_skb); 1556 } else { 1557 __kfree_skb(opt_skb); 1558 opt_skb = xchg(&np->pktoptions, NULL); 1559 } 1560 } 1561 1562 consume_skb(opt_skb); 1563 return 0; 1564 } 1565 1566 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, 1567 const struct tcphdr *th) 1568 { 1569 /* This is tricky: we move IP6CB at its correct location into 1570 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because 1571 * _decode_session6() uses IP6CB(). 1572 * barrier() makes sure compiler won't play aliasing games. 1573 */ 1574 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), 1575 sizeof(struct inet6_skb_parm)); 1576 barrier(); 1577 1578 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1579 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1580 skb->len - th->doff*4); 1581 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1582 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1583 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1584 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr); 1585 TCP_SKB_CB(skb)->sacked = 0; 1586 TCP_SKB_CB(skb)->has_rxtstamp = 1587 skb->tstamp || skb_hwtstamps(skb)->hwtstamp; 1588 } 1589 1590 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) 1591 { 1592 enum skb_drop_reason drop_reason; 1593 int sdif = inet6_sdif(skb); 1594 int dif = inet6_iif(skb); 1595 const struct tcphdr *th; 1596 const struct ipv6hdr *hdr; 1597 bool refcounted; 1598 struct sock *sk; 1599 int ret; 1600 struct net *net = dev_net(skb->dev); 1601 1602 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 1603 if (skb->pkt_type != PACKET_HOST) 1604 goto discard_it; 1605 1606 /* 1607 * Count it even if it's bad. 1608 */ 1609 __TCP_INC_STATS(net, TCP_MIB_INSEGS); 1610 1611 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1612 goto discard_it; 1613 1614 th = (const struct tcphdr *)skb->data; 1615 1616 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) { 1617 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1618 goto bad_packet; 1619 } 1620 if (!pskb_may_pull(skb, th->doff*4)) 1621 goto discard_it; 1622 1623 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) 1624 goto csum_error; 1625 1626 th = (const struct tcphdr *)skb->data; 1627 hdr = ipv6_hdr(skb); 1628 1629 lookup: 1630 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th), 1631 th->source, th->dest, inet6_iif(skb), sdif, 1632 &refcounted); 1633 if (!sk) 1634 goto no_tcp_socket; 1635 1636 process: 1637 if (sk->sk_state == TCP_TIME_WAIT) 1638 goto do_time_wait; 1639 1640 if (sk->sk_state == TCP_NEW_SYN_RECV) { 1641 struct request_sock *req = inet_reqsk(sk); 1642 bool req_stolen = false; 1643 struct sock *nsk; 1644 1645 sk = req->rsk_listener; 1646 drop_reason = tcp_inbound_md5_hash(sk, skb, 1647 &hdr->saddr, &hdr->daddr, 1648 AF_INET6, dif, sdif); 1649 if (drop_reason) { 1650 sk_drops_add(sk, skb); 1651 reqsk_put(req); 1652 goto discard_it; 1653 } 1654 if (tcp_checksum_complete(skb)) { 1655 reqsk_put(req); 1656 goto csum_error; 1657 } 1658 if (unlikely(sk->sk_state != TCP_LISTEN)) { 1659 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb); 1660 if (!nsk) { 1661 inet_csk_reqsk_queue_drop_and_put(sk, req); 1662 goto lookup; 1663 } 1664 sk = nsk; 1665 /* reuseport_migrate_sock() has already held one sk_refcnt 1666 * before returning. 1667 */ 1668 } else { 1669 sock_hold(sk); 1670 } 1671 refcounted = true; 1672 nsk = NULL; 1673 if (!tcp_filter(sk, skb)) { 1674 th = (const struct tcphdr *)skb->data; 1675 hdr = ipv6_hdr(skb); 1676 tcp_v6_fill_cb(skb, hdr, th); 1677 nsk = tcp_check_req(sk, skb, req, false, &req_stolen); 1678 } else { 1679 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1680 } 1681 if (!nsk) { 1682 reqsk_put(req); 1683 if (req_stolen) { 1684 /* Another cpu got exclusive access to req 1685 * and created a full blown socket. 1686 * Try to feed this packet to this socket 1687 * instead of discarding it. 1688 */ 1689 tcp_v6_restore_cb(skb); 1690 sock_put(sk); 1691 goto lookup; 1692 } 1693 goto discard_and_relse; 1694 } 1695 if (nsk == sk) { 1696 reqsk_put(req); 1697 tcp_v6_restore_cb(skb); 1698 } else if (tcp_child_process(sk, nsk, skb)) { 1699 tcp_v6_send_reset(nsk, skb); 1700 goto discard_and_relse; 1701 } else { 1702 sock_put(sk); 1703 return 0; 1704 } 1705 } 1706 1707 if (static_branch_unlikely(&ip6_min_hopcount)) { 1708 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ 1709 if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) { 1710 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 1711 drop_reason = SKB_DROP_REASON_TCP_MINTTL; 1712 goto discard_and_relse; 1713 } 1714 } 1715 1716 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 1717 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1718 goto discard_and_relse; 1719 } 1720 1721 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr, 1722 AF_INET6, dif, sdif); 1723 if (drop_reason) 1724 goto discard_and_relse; 1725 1726 if (tcp_filter(sk, skb)) { 1727 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1728 goto discard_and_relse; 1729 } 1730 th = (const struct tcphdr *)skb->data; 1731 hdr = ipv6_hdr(skb); 1732 tcp_v6_fill_cb(skb, hdr, th); 1733 1734 skb->dev = NULL; 1735 1736 if (sk->sk_state == TCP_LISTEN) { 1737 ret = tcp_v6_do_rcv(sk, skb); 1738 goto put_and_return; 1739 } 1740 1741 sk_incoming_cpu_update(sk); 1742 1743 bh_lock_sock_nested(sk); 1744 tcp_segs_in(tcp_sk(sk), skb); 1745 ret = 0; 1746 if (!sock_owned_by_user(sk)) { 1747 ret = tcp_v6_do_rcv(sk, skb); 1748 } else { 1749 if (tcp_add_backlog(sk, skb, &drop_reason)) 1750 goto discard_and_relse; 1751 } 1752 bh_unlock_sock(sk); 1753 put_and_return: 1754 if (refcounted) 1755 sock_put(sk); 1756 return ret ? -1 : 0; 1757 1758 no_tcp_socket: 1759 drop_reason = SKB_DROP_REASON_NO_SOCKET; 1760 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1761 goto discard_it; 1762 1763 tcp_v6_fill_cb(skb, hdr, th); 1764 1765 if (tcp_checksum_complete(skb)) { 1766 csum_error: 1767 drop_reason = SKB_DROP_REASON_TCP_CSUM; 1768 trace_tcp_bad_csum(skb); 1769 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); 1770 bad_packet: 1771 __TCP_INC_STATS(net, TCP_MIB_INERRS); 1772 } else { 1773 tcp_v6_send_reset(NULL, skb); 1774 } 1775 1776 discard_it: 1777 SKB_DR_OR(drop_reason, NOT_SPECIFIED); 1778 kfree_skb_reason(skb, drop_reason); 1779 return 0; 1780 1781 discard_and_relse: 1782 sk_drops_add(sk, skb); 1783 if (refcounted) 1784 sock_put(sk); 1785 goto discard_it; 1786 1787 do_time_wait: 1788 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1789 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1790 inet_twsk_put(inet_twsk(sk)); 1791 goto discard_it; 1792 } 1793 1794 tcp_v6_fill_cb(skb, hdr, th); 1795 1796 if (tcp_checksum_complete(skb)) { 1797 inet_twsk_put(inet_twsk(sk)); 1798 goto csum_error; 1799 } 1800 1801 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1802 case TCP_TW_SYN: 1803 { 1804 struct sock *sk2; 1805 1806 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1807 skb, __tcp_hdrlen(th), 1808 &ipv6_hdr(skb)->saddr, th->source, 1809 &ipv6_hdr(skb)->daddr, 1810 ntohs(th->dest), 1811 tcp_v6_iif_l3_slave(skb), 1812 sdif); 1813 if (sk2) { 1814 struct inet_timewait_sock *tw = inet_twsk(sk); 1815 inet_twsk_deschedule_put(tw); 1816 sk = sk2; 1817 tcp_v6_restore_cb(skb); 1818 refcounted = false; 1819 goto process; 1820 } 1821 } 1822 /* to ACK */ 1823 fallthrough; 1824 case TCP_TW_ACK: 1825 tcp_v6_timewait_ack(sk, skb); 1826 break; 1827 case TCP_TW_RST: 1828 tcp_v6_send_reset(sk, skb); 1829 inet_twsk_deschedule_put(inet_twsk(sk)); 1830 goto discard_it; 1831 case TCP_TW_SUCCESS: 1832 ; 1833 } 1834 goto discard_it; 1835 } 1836 1837 void tcp_v6_early_demux(struct sk_buff *skb) 1838 { 1839 struct net *net = dev_net(skb->dev); 1840 const struct ipv6hdr *hdr; 1841 const struct tcphdr *th; 1842 struct sock *sk; 1843 1844 if (skb->pkt_type != PACKET_HOST) 1845 return; 1846 1847 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1848 return; 1849 1850 hdr = ipv6_hdr(skb); 1851 th = tcp_hdr(skb); 1852 1853 if (th->doff < sizeof(struct tcphdr) / 4) 1854 return; 1855 1856 /* Note : We use inet6_iif() here, not tcp_v6_iif() */ 1857 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 1858 &hdr->saddr, th->source, 1859 &hdr->daddr, ntohs(th->dest), 1860 inet6_iif(skb), inet6_sdif(skb)); 1861 if (sk) { 1862 skb->sk = sk; 1863 skb->destructor = sock_edemux; 1864 if (sk_fullsock(sk)) { 1865 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst); 1866 1867 if (dst) 1868 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1869 if (dst && 1870 sk->sk_rx_dst_ifindex == skb->skb_iif) 1871 skb_dst_set_noref(skb, dst); 1872 } 1873 } 1874 } 1875 1876 static struct timewait_sock_ops tcp6_timewait_sock_ops = { 1877 .twsk_obj_size = sizeof(struct tcp6_timewait_sock), 1878 .twsk_unique = tcp_twsk_unique, 1879 .twsk_destructor = tcp_twsk_destructor, 1880 }; 1881 1882 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb) 1883 { 1884 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr); 1885 } 1886 1887 const struct inet_connection_sock_af_ops ipv6_specific = { 1888 .queue_xmit = inet6_csk_xmit, 1889 .send_check = tcp_v6_send_check, 1890 .rebuild_header = inet6_sk_rebuild_header, 1891 .sk_rx_dst_set = inet6_sk_rx_dst_set, 1892 .conn_request = tcp_v6_conn_request, 1893 .syn_recv_sock = tcp_v6_syn_recv_sock, 1894 .net_header_len = sizeof(struct ipv6hdr), 1895 .net_frag_header_len = sizeof(struct frag_hdr), 1896 .setsockopt = ipv6_setsockopt, 1897 .getsockopt = ipv6_getsockopt, 1898 .addr2sockaddr = inet6_csk_addr2sockaddr, 1899 .sockaddr_len = sizeof(struct sockaddr_in6), 1900 .mtu_reduced = tcp_v6_mtu_reduced, 1901 }; 1902 1903 #ifdef CONFIG_TCP_MD5SIG 1904 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { 1905 .md5_lookup = tcp_v6_md5_lookup, 1906 .calc_md5_hash = tcp_v6_md5_hash_skb, 1907 .md5_parse = tcp_v6_parse_md5_keys, 1908 }; 1909 #endif 1910 1911 /* 1912 * TCP over IPv4 via INET6 API 1913 */ 1914 static const struct inet_connection_sock_af_ops ipv6_mapped = { 1915 .queue_xmit = ip_queue_xmit, 1916 .send_check = tcp_v4_send_check, 1917 .rebuild_header = inet_sk_rebuild_header, 1918 .sk_rx_dst_set = inet_sk_rx_dst_set, 1919 .conn_request = tcp_v6_conn_request, 1920 .syn_recv_sock = tcp_v6_syn_recv_sock, 1921 .net_header_len = sizeof(struct iphdr), 1922 .setsockopt = ipv6_setsockopt, 1923 .getsockopt = ipv6_getsockopt, 1924 .addr2sockaddr = inet6_csk_addr2sockaddr, 1925 .sockaddr_len = sizeof(struct sockaddr_in6), 1926 .mtu_reduced = tcp_v4_mtu_reduced, 1927 }; 1928 1929 #ifdef CONFIG_TCP_MD5SIG 1930 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { 1931 .md5_lookup = tcp_v4_md5_lookup, 1932 .calc_md5_hash = tcp_v4_md5_hash_skb, 1933 .md5_parse = tcp_v6_parse_md5_keys, 1934 }; 1935 #endif 1936 1937 /* NOTE: A lot of things set to zero explicitly by call to 1938 * sk_alloc() so need not be done here. 1939 */ 1940 static int tcp_v6_init_sock(struct sock *sk) 1941 { 1942 struct inet_connection_sock *icsk = inet_csk(sk); 1943 1944 tcp_init_sock(sk); 1945 1946 icsk->icsk_af_ops = &ipv6_specific; 1947 1948 #ifdef CONFIG_TCP_MD5SIG 1949 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; 1950 #endif 1951 1952 return 0; 1953 } 1954 1955 #ifdef CONFIG_PROC_FS 1956 /* Proc filesystem TCPv6 sock list dumping. */ 1957 static void get_openreq6(struct seq_file *seq, 1958 const struct request_sock *req, int i) 1959 { 1960 long ttd = req->rsk_timer.expires - jiffies; 1961 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr; 1962 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr; 1963 1964 if (ttd < 0) 1965 ttd = 0; 1966 1967 seq_printf(seq, 1968 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1969 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", 1970 i, 1971 src->s6_addr32[0], src->s6_addr32[1], 1972 src->s6_addr32[2], src->s6_addr32[3], 1973 inet_rsk(req)->ir_num, 1974 dest->s6_addr32[0], dest->s6_addr32[1], 1975 dest->s6_addr32[2], dest->s6_addr32[3], 1976 ntohs(inet_rsk(req)->ir_rmt_port), 1977 TCP_SYN_RECV, 1978 0, 0, /* could print option size, but that is af dependent. */ 1979 1, /* timers active (only the expire timer) */ 1980 jiffies_to_clock_t(ttd), 1981 req->num_timeout, 1982 from_kuid_munged(seq_user_ns(seq), 1983 sock_i_uid(req->rsk_listener)), 1984 0, /* non standard timer */ 1985 0, /* open_requests have no inode */ 1986 0, req); 1987 } 1988 1989 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) 1990 { 1991 const struct in6_addr *dest, *src; 1992 __u16 destp, srcp; 1993 int timer_active; 1994 unsigned long timer_expires; 1995 const struct inet_sock *inet = inet_sk(sp); 1996 const struct tcp_sock *tp = tcp_sk(sp); 1997 const struct inet_connection_sock *icsk = inet_csk(sp); 1998 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; 1999 int rx_queue; 2000 int state; 2001 2002 dest = &sp->sk_v6_daddr; 2003 src = &sp->sk_v6_rcv_saddr; 2004 destp = ntohs(inet->inet_dport); 2005 srcp = ntohs(inet->inet_sport); 2006 2007 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 2008 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || 2009 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 2010 timer_active = 1; 2011 timer_expires = icsk->icsk_timeout; 2012 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 2013 timer_active = 4; 2014 timer_expires = icsk->icsk_timeout; 2015 } else if (timer_pending(&sp->sk_timer)) { 2016 timer_active = 2; 2017 timer_expires = sp->sk_timer.expires; 2018 } else { 2019 timer_active = 0; 2020 timer_expires = jiffies; 2021 } 2022 2023 state = inet_sk_state_load(sp); 2024 if (state == TCP_LISTEN) 2025 rx_queue = READ_ONCE(sp->sk_ack_backlog); 2026 else 2027 /* Because we don't lock the socket, 2028 * we might find a transient negative value. 2029 */ 2030 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) - 2031 READ_ONCE(tp->copied_seq), 0); 2032 2033 seq_printf(seq, 2034 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2035 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", 2036 i, 2037 src->s6_addr32[0], src->s6_addr32[1], 2038 src->s6_addr32[2], src->s6_addr32[3], srcp, 2039 dest->s6_addr32[0], dest->s6_addr32[1], 2040 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2041 state, 2042 READ_ONCE(tp->write_seq) - tp->snd_una, 2043 rx_queue, 2044 timer_active, 2045 jiffies_delta_to_clock_t(timer_expires - jiffies), 2046 icsk->icsk_retransmits, 2047 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)), 2048 icsk->icsk_probes_out, 2049 sock_i_ino(sp), 2050 refcount_read(&sp->sk_refcnt), sp, 2051 jiffies_to_clock_t(icsk->icsk_rto), 2052 jiffies_to_clock_t(icsk->icsk_ack.ato), 2053 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp), 2054 tcp_snd_cwnd(tp), 2055 state == TCP_LISTEN ? 2056 fastopenq->max_qlen : 2057 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) 2058 ); 2059 } 2060 2061 static void get_timewait6_sock(struct seq_file *seq, 2062 struct inet_timewait_sock *tw, int i) 2063 { 2064 long delta = tw->tw_timer.expires - jiffies; 2065 const struct in6_addr *dest, *src; 2066 __u16 destp, srcp; 2067 2068 dest = &tw->tw_v6_daddr; 2069 src = &tw->tw_v6_rcv_saddr; 2070 destp = ntohs(tw->tw_dport); 2071 srcp = ntohs(tw->tw_sport); 2072 2073 seq_printf(seq, 2074 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2075 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", 2076 i, 2077 src->s6_addr32[0], src->s6_addr32[1], 2078 src->s6_addr32[2], src->s6_addr32[3], srcp, 2079 dest->s6_addr32[0], dest->s6_addr32[1], 2080 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2081 tw->tw_substate, 0, 0, 2082 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2083 refcount_read(&tw->tw_refcnt), tw); 2084 } 2085 2086 static int tcp6_seq_show(struct seq_file *seq, void *v) 2087 { 2088 struct tcp_iter_state *st; 2089 struct sock *sk = v; 2090 2091 if (v == SEQ_START_TOKEN) { 2092 seq_puts(seq, 2093 " sl " 2094 "local_address " 2095 "remote_address " 2096 "st tx_queue rx_queue tr tm->when retrnsmt" 2097 " uid timeout inode\n"); 2098 goto out; 2099 } 2100 st = seq->private; 2101 2102 if (sk->sk_state == TCP_TIME_WAIT) 2103 get_timewait6_sock(seq, v, st->num); 2104 else if (sk->sk_state == TCP_NEW_SYN_RECV) 2105 get_openreq6(seq, v, st->num); 2106 else 2107 get_tcp6_sock(seq, v, st->num); 2108 out: 2109 return 0; 2110 } 2111 2112 static const struct seq_operations tcp6_seq_ops = { 2113 .show = tcp6_seq_show, 2114 .start = tcp_seq_start, 2115 .next = tcp_seq_next, 2116 .stop = tcp_seq_stop, 2117 }; 2118 2119 static struct tcp_seq_afinfo tcp6_seq_afinfo = { 2120 .family = AF_INET6, 2121 }; 2122 2123 int __net_init tcp6_proc_init(struct net *net) 2124 { 2125 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops, 2126 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo)) 2127 return -ENOMEM; 2128 return 0; 2129 } 2130 2131 void tcp6_proc_exit(struct net *net) 2132 { 2133 remove_proc_entry("tcp6", net->proc_net); 2134 } 2135 #endif 2136 2137 struct proto tcpv6_prot = { 2138 .name = "TCPv6", 2139 .owner = THIS_MODULE, 2140 .close = tcp_close, 2141 .pre_connect = tcp_v6_pre_connect, 2142 .connect = tcp_v6_connect, 2143 .disconnect = tcp_disconnect, 2144 .accept = inet_csk_accept, 2145 .ioctl = tcp_ioctl, 2146 .init = tcp_v6_init_sock, 2147 .destroy = tcp_v4_destroy_sock, 2148 .shutdown = tcp_shutdown, 2149 .setsockopt = tcp_setsockopt, 2150 .getsockopt = tcp_getsockopt, 2151 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt, 2152 .keepalive = tcp_set_keepalive, 2153 .recvmsg = tcp_recvmsg, 2154 .sendmsg = tcp_sendmsg, 2155 .sendpage = tcp_sendpage, 2156 .backlog_rcv = tcp_v6_do_rcv, 2157 .release_cb = tcp_release_cb, 2158 .hash = inet6_hash, 2159 .unhash = inet_unhash, 2160 .get_port = inet_csk_get_port, 2161 .put_port = inet_put_port, 2162 #ifdef CONFIG_BPF_SYSCALL 2163 .psock_update_sk_prot = tcp_bpf_update_proto, 2164 #endif 2165 .enter_memory_pressure = tcp_enter_memory_pressure, 2166 .leave_memory_pressure = tcp_leave_memory_pressure, 2167 .stream_memory_free = tcp_stream_memory_free, 2168 .sockets_allocated = &tcp_sockets_allocated, 2169 2170 .memory_allocated = &tcp_memory_allocated, 2171 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc, 2172 2173 .memory_pressure = &tcp_memory_pressure, 2174 .orphan_count = &tcp_orphan_count, 2175 .sysctl_mem = sysctl_tcp_mem, 2176 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), 2177 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), 2178 .max_header = MAX_TCP_HEADER, 2179 .obj_size = sizeof(struct tcp6_sock), 2180 .slab_flags = SLAB_TYPESAFE_BY_RCU, 2181 .twsk_prot = &tcp6_timewait_sock_ops, 2182 .rsk_prot = &tcp6_request_sock_ops, 2183 .h.hashinfo = NULL, 2184 .no_autobind = true, 2185 .diag_destroy = tcp_abort, 2186 }; 2187 EXPORT_SYMBOL_GPL(tcpv6_prot); 2188 2189 static const struct inet6_protocol tcpv6_protocol = { 2190 .handler = tcp_v6_rcv, 2191 .err_handler = tcp_v6_err, 2192 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 2193 }; 2194 2195 static struct inet_protosw tcpv6_protosw = { 2196 .type = SOCK_STREAM, 2197 .protocol = IPPROTO_TCP, 2198 .prot = &tcpv6_prot, 2199 .ops = &inet6_stream_ops, 2200 .flags = INET_PROTOSW_PERMANENT | 2201 INET_PROTOSW_ICSK, 2202 }; 2203 2204 static int __net_init tcpv6_net_init(struct net *net) 2205 { 2206 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, 2207 SOCK_RAW, IPPROTO_TCP, net); 2208 } 2209 2210 static void __net_exit tcpv6_net_exit(struct net *net) 2211 { 2212 inet_ctl_sock_destroy(net->ipv6.tcp_sk); 2213 } 2214 2215 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list) 2216 { 2217 tcp_twsk_purge(net_exit_list, AF_INET6); 2218 } 2219 2220 static struct pernet_operations tcpv6_net_ops = { 2221 .init = tcpv6_net_init, 2222 .exit = tcpv6_net_exit, 2223 .exit_batch = tcpv6_net_exit_batch, 2224 }; 2225 2226 int __init tcpv6_init(void) 2227 { 2228 int ret; 2229 2230 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); 2231 if (ret) 2232 goto out; 2233 2234 /* register inet6 protocol */ 2235 ret = inet6_register_protosw(&tcpv6_protosw); 2236 if (ret) 2237 goto out_tcpv6_protocol; 2238 2239 ret = register_pernet_subsys(&tcpv6_net_ops); 2240 if (ret) 2241 goto out_tcpv6_protosw; 2242 2243 ret = mptcpv6_init(); 2244 if (ret) 2245 goto out_tcpv6_pernet_subsys; 2246 2247 out: 2248 return ret; 2249 2250 out_tcpv6_pernet_subsys: 2251 unregister_pernet_subsys(&tcpv6_net_ops); 2252 out_tcpv6_protosw: 2253 inet6_unregister_protosw(&tcpv6_protosw); 2254 out_tcpv6_protocol: 2255 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2256 goto out; 2257 } 2258 2259 void tcpv6_exit(void) 2260 { 2261 unregister_pernet_subsys(&tcpv6_net_ops); 2262 inet6_unregister_protosw(&tcpv6_protosw); 2263 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2264 } 2265