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 #define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \ 97 struct tcp6_sock, tcp)->inet6) 98 99 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) 100 { 101 struct dst_entry *dst = skb_dst(skb); 102 103 if (dst && dst_hold_safe(dst)) { 104 const struct rt6_info *rt = (const struct rt6_info *)dst; 105 106 rcu_assign_pointer(sk->sk_rx_dst, dst); 107 sk->sk_rx_dst_ifindex = skb->skb_iif; 108 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 109 } 110 } 111 112 static u32 tcp_v6_init_seq(const struct sk_buff *skb) 113 { 114 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32, 115 ipv6_hdr(skb)->saddr.s6_addr32, 116 tcp_hdr(skb)->dest, 117 tcp_hdr(skb)->source); 118 } 119 120 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb) 121 { 122 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32, 123 ipv6_hdr(skb)->saddr.s6_addr32); 124 } 125 126 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 127 int addr_len) 128 { 129 /* This check is replicated from tcp_v6_connect() and intended to 130 * prevent BPF program called below from accessing bytes that are out 131 * of the bound specified by user in addr_len. 132 */ 133 if (addr_len < SIN6_LEN_RFC2133) 134 return -EINVAL; 135 136 sock_owned_by_me(sk); 137 138 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len); 139 } 140 141 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr, 142 int addr_len) 143 { 144 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr; 145 struct inet_connection_sock *icsk = inet_csk(sk); 146 struct in6_addr *saddr = NULL, *final_p, final; 147 struct inet_timewait_death_row *tcp_death_row; 148 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 149 struct inet_sock *inet = inet_sk(sk); 150 struct tcp_sock *tp = tcp_sk(sk); 151 struct net *net = sock_net(sk); 152 struct ipv6_txoptions *opt; 153 struct dst_entry *dst; 154 struct flowi6 fl6; 155 int addr_type; 156 int err; 157 158 if (addr_len < SIN6_LEN_RFC2133) 159 return -EINVAL; 160 161 if (usin->sin6_family != AF_INET6) 162 return -EAFNOSUPPORT; 163 164 memset(&fl6, 0, sizeof(fl6)); 165 166 if (inet6_test_bit(SNDFLOW, sk)) { 167 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; 168 IP6_ECN_flow_init(fl6.flowlabel); 169 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 170 struct ip6_flowlabel *flowlabel; 171 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 172 if (IS_ERR(flowlabel)) 173 return -EINVAL; 174 fl6_sock_release(flowlabel); 175 } 176 } 177 178 /* 179 * connect() to INADDR_ANY means loopback (BSD'ism). 180 */ 181 182 if (ipv6_addr_any(&usin->sin6_addr)) { 183 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 184 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 185 &usin->sin6_addr); 186 else 187 usin->sin6_addr = in6addr_loopback; 188 } 189 190 addr_type = ipv6_addr_type(&usin->sin6_addr); 191 192 if (addr_type & IPV6_ADDR_MULTICAST) 193 return -ENETUNREACH; 194 195 if (addr_type&IPV6_ADDR_LINKLOCAL) { 196 if (addr_len >= sizeof(struct sockaddr_in6) && 197 usin->sin6_scope_id) { 198 /* If interface is set while binding, indices 199 * must coincide. 200 */ 201 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id)) 202 return -EINVAL; 203 204 sk->sk_bound_dev_if = usin->sin6_scope_id; 205 } 206 207 /* Connect to link-local address requires an interface */ 208 if (!sk->sk_bound_dev_if) 209 return -EINVAL; 210 } 211 212 if (tp->rx_opt.ts_recent_stamp && 213 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) { 214 tp->rx_opt.ts_recent = 0; 215 tp->rx_opt.ts_recent_stamp = 0; 216 WRITE_ONCE(tp->write_seq, 0); 217 } 218 219 sk->sk_v6_daddr = usin->sin6_addr; 220 np->flow_label = fl6.flowlabel; 221 222 /* 223 * TCP over IPv4 224 */ 225 226 if (addr_type & IPV6_ADDR_MAPPED) { 227 u32 exthdrlen = icsk->icsk_ext_hdr_len; 228 struct sockaddr_in sin; 229 230 if (ipv6_only_sock(sk)) 231 return -ENETUNREACH; 232 233 sin.sin_family = AF_INET; 234 sin.sin_port = usin->sin6_port; 235 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 236 237 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ 238 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped); 239 if (sk_is_mptcp(sk)) 240 mptcpv6_handle_mapped(sk, true); 241 sk->sk_backlog_rcv = tcp_v4_do_rcv; 242 #ifdef CONFIG_TCP_MD5SIG 243 tp->af_specific = &tcp_sock_ipv6_mapped_specific; 244 #endif 245 246 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 247 248 if (err) { 249 icsk->icsk_ext_hdr_len = exthdrlen; 250 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ 251 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific); 252 if (sk_is_mptcp(sk)) 253 mptcpv6_handle_mapped(sk, false); 254 sk->sk_backlog_rcv = tcp_v6_do_rcv; 255 #ifdef CONFIG_TCP_MD5SIG 256 tp->af_specific = &tcp_sock_ipv6_specific; 257 #endif 258 goto failure; 259 } 260 np->saddr = sk->sk_v6_rcv_saddr; 261 262 return err; 263 } 264 265 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 266 saddr = &sk->sk_v6_rcv_saddr; 267 268 fl6.flowi6_proto = IPPROTO_TCP; 269 fl6.daddr = sk->sk_v6_daddr; 270 fl6.saddr = saddr ? *saddr : np->saddr; 271 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label); 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(net, sk, &fl6, final_p); 284 if (IS_ERR(dst)) { 285 err = PTR_ERR(dst); 286 goto failure; 287 } 288 289 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; 290 291 if (!saddr) { 292 saddr = &fl6.saddr; 293 294 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6); 295 if (err) 296 goto failure; 297 } 298 299 /* set the source address */ 300 np->saddr = *saddr; 301 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 302 303 sk->sk_gso_type = SKB_GSO_TCPV6; 304 ip6_dst_store(sk, dst, NULL, NULL); 305 306 icsk->icsk_ext_hdr_len = 0; 307 if (opt) 308 icsk->icsk_ext_hdr_len = opt->opt_flen + 309 opt->opt_nflen; 310 311 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 312 313 inet->inet_dport = usin->sin6_port; 314 315 tcp_set_state(sk, TCP_SYN_SENT); 316 err = inet6_hash_connect(tcp_death_row, sk); 317 if (err) 318 goto late_failure; 319 320 sk_set_txhash(sk); 321 322 if (likely(!tp->repair)) { 323 if (!tp->write_seq) 324 WRITE_ONCE(tp->write_seq, 325 secure_tcpv6_seq(np->saddr.s6_addr32, 326 sk->sk_v6_daddr.s6_addr32, 327 inet->inet_sport, 328 inet->inet_dport)); 329 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32, 330 sk->sk_v6_daddr.s6_addr32); 331 } 332 333 if (tcp_fastopen_defer_connect(sk, &err)) 334 return err; 335 if (err) 336 goto late_failure; 337 338 err = tcp_connect(sk); 339 if (err) 340 goto late_failure; 341 342 return 0; 343 344 late_failure: 345 tcp_set_state(sk, TCP_CLOSE); 346 inet_bhash2_reset_saddr(sk); 347 failure: 348 inet->inet_dport = 0; 349 sk->sk_route_caps = 0; 350 return err; 351 } 352 353 static void tcp_v6_mtu_reduced(struct sock *sk) 354 { 355 struct dst_entry *dst; 356 u32 mtu; 357 358 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) 359 return; 360 361 mtu = READ_ONCE(tcp_sk(sk)->mtu_info); 362 363 /* Drop requests trying to increase our current mss. 364 * Check done in __ip6_rt_update_pmtu() is too late. 365 */ 366 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache) 367 return; 368 369 dst = inet6_csk_update_pmtu(sk, mtu); 370 if (!dst) 371 return; 372 373 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 374 tcp_sync_mss(sk, dst_mtu(dst)); 375 tcp_simple_retransmit(sk); 376 } 377 } 378 379 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 380 u8 type, u8 code, int offset, __be32 info) 381 { 382 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 383 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); 384 struct net *net = dev_net(skb->dev); 385 struct request_sock *fastopen; 386 struct ipv6_pinfo *np; 387 struct tcp_sock *tp; 388 __u32 seq, snd_una; 389 struct sock *sk; 390 bool fatal; 391 int err; 392 393 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 394 &hdr->daddr, th->dest, 395 &hdr->saddr, ntohs(th->source), 396 skb->dev->ifindex, inet6_sdif(skb)); 397 398 if (!sk) { 399 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 400 ICMP6_MIB_INERRORS); 401 return -ENOENT; 402 } 403 404 if (sk->sk_state == TCP_TIME_WAIT) { 405 inet_twsk_put(inet_twsk(sk)); 406 return 0; 407 } 408 seq = ntohl(th->seq); 409 fatal = icmpv6_err_convert(type, code, &err); 410 if (sk->sk_state == TCP_NEW_SYN_RECV) { 411 tcp_req_err(sk, seq, fatal); 412 return 0; 413 } 414 415 bh_lock_sock(sk); 416 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) 417 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 418 419 if (sk->sk_state == TCP_CLOSE) 420 goto out; 421 422 if (static_branch_unlikely(&ip6_min_hopcount)) { 423 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ 424 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) { 425 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 426 goto out; 427 } 428 } 429 430 tp = tcp_sk(sk); 431 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ 432 fastopen = rcu_dereference(tp->fastopen_rsk); 433 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; 434 if (sk->sk_state != TCP_LISTEN && 435 !between(seq, snd_una, tp->snd_nxt)) { 436 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 437 goto out; 438 } 439 440 np = tcp_inet6_sk(sk); 441 442 if (type == NDISC_REDIRECT) { 443 if (!sock_owned_by_user(sk)) { 444 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie); 445 446 if (dst) 447 dst->ops->redirect(dst, sk, skb); 448 } 449 goto out; 450 } 451 452 if (type == ICMPV6_PKT_TOOBIG) { 453 u32 mtu = ntohl(info); 454 455 /* We are not interested in TCP_LISTEN and open_requests 456 * (SYN-ACKs send out by Linux are always <576bytes so 457 * they should go through unfragmented). 458 */ 459 if (sk->sk_state == TCP_LISTEN) 460 goto out; 461 462 if (!ip6_sk_accept_pmtu(sk)) 463 goto out; 464 465 if (mtu < IPV6_MIN_MTU) 466 goto out; 467 468 WRITE_ONCE(tp->mtu_info, mtu); 469 470 if (!sock_owned_by_user(sk)) 471 tcp_v6_mtu_reduced(sk); 472 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, 473 &sk->sk_tsq_flags)) 474 sock_hold(sk); 475 goto out; 476 } 477 478 479 /* Might be for an request_sock */ 480 switch (sk->sk_state) { 481 case TCP_SYN_SENT: 482 case TCP_SYN_RECV: 483 /* Only in fast or simultaneous open. If a fast open socket is 484 * already accepted it is treated as a connected one below. 485 */ 486 if (fastopen && !fastopen->sk) 487 break; 488 489 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th); 490 491 if (!sock_owned_by_user(sk)) { 492 WRITE_ONCE(sk->sk_err, err); 493 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ 494 495 tcp_done(sk); 496 } else { 497 WRITE_ONCE(sk->sk_err_soft, err); 498 } 499 goto out; 500 case TCP_LISTEN: 501 break; 502 default: 503 /* check if this ICMP message allows revert of backoff. 504 * (see RFC 6069) 505 */ 506 if (!fastopen && type == ICMPV6_DEST_UNREACH && 507 code == ICMPV6_NOROUTE) 508 tcp_ld_RTO_revert(sk, seq); 509 } 510 511 if (!sock_owned_by_user(sk) && inet6_test_bit(RECVERR6, sk)) { 512 WRITE_ONCE(sk->sk_err, err); 513 sk_error_report(sk); 514 } else { 515 WRITE_ONCE(sk->sk_err_soft, err); 516 } 517 out: 518 bh_unlock_sock(sk); 519 sock_put(sk); 520 return 0; 521 } 522 523 524 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, 525 struct flowi *fl, 526 struct request_sock *req, 527 struct tcp_fastopen_cookie *foc, 528 enum tcp_synack_type synack_type, 529 struct sk_buff *syn_skb) 530 { 531 struct inet_request_sock *ireq = inet_rsk(req); 532 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 533 struct ipv6_txoptions *opt; 534 struct flowi6 *fl6 = &fl->u.ip6; 535 struct sk_buff *skb; 536 int err = -ENOMEM; 537 u8 tclass; 538 539 /* First, grab a route. */ 540 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req, 541 IPPROTO_TCP)) == NULL) 542 goto done; 543 544 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb); 545 546 if (skb) { 547 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, 548 &ireq->ir_v6_rmt_addr); 549 550 fl6->daddr = ireq->ir_v6_rmt_addr; 551 if (inet6_test_bit(REPFLOW, sk) && ireq->pktopts) 552 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); 553 554 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ? 555 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) | 556 (np->tclass & INET_ECN_MASK) : 557 np->tclass; 558 559 if (!INET_ECN_is_capable(tclass) && 560 tcp_bpf_ca_needs_ecn((struct sock *)req)) 561 tclass |= INET_ECN_ECT_0; 562 563 rcu_read_lock(); 564 opt = ireq->ipv6_opt; 565 if (!opt) 566 opt = rcu_dereference(np->opt); 567 err = ip6_xmit(sk, skb, fl6, skb->mark ? : READ_ONCE(sk->sk_mark), 568 opt, tclass, READ_ONCE(sk->sk_priority)); 569 rcu_read_unlock(); 570 err = net_xmit_eval(err); 571 } 572 573 done: 574 return err; 575 } 576 577 578 static void tcp_v6_reqsk_destructor(struct request_sock *req) 579 { 580 kfree(inet_rsk(req)->ipv6_opt); 581 consume_skb(inet_rsk(req)->pktopts); 582 } 583 584 #ifdef CONFIG_TCP_MD5SIG 585 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, 586 const struct in6_addr *addr, 587 int l3index) 588 { 589 return tcp_md5_do_lookup(sk, l3index, 590 (union tcp_md5_addr *)addr, AF_INET6); 591 } 592 593 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk, 594 const struct sock *addr_sk) 595 { 596 int l3index; 597 598 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 599 addr_sk->sk_bound_dev_if); 600 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr, 601 l3index); 602 } 603 604 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname, 605 sockptr_t optval, int optlen) 606 { 607 struct tcp_md5sig cmd; 608 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; 609 int l3index = 0; 610 u8 prefixlen; 611 u8 flags; 612 613 if (optlen < sizeof(cmd)) 614 return -EINVAL; 615 616 if (copy_from_sockptr(&cmd, optval, sizeof(cmd))) 617 return -EFAULT; 618 619 if (sin6->sin6_family != AF_INET6) 620 return -EINVAL; 621 622 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX; 623 624 if (optname == TCP_MD5SIG_EXT && 625 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { 626 prefixlen = cmd.tcpm_prefixlen; 627 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) && 628 prefixlen > 32)) 629 return -EINVAL; 630 } else { 631 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128; 632 } 633 634 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex && 635 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) { 636 struct net_device *dev; 637 638 rcu_read_lock(); 639 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex); 640 if (dev && netif_is_l3_master(dev)) 641 l3index = dev->ifindex; 642 rcu_read_unlock(); 643 644 /* ok to reference set/not set outside of rcu; 645 * right now device MUST be an L3 master 646 */ 647 if (!dev || !l3index) 648 return -EINVAL; 649 } 650 651 if (!cmd.tcpm_keylen) { 652 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 653 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 654 AF_INET, prefixlen, 655 l3index, flags); 656 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 657 AF_INET6, prefixlen, l3index, flags); 658 } 659 660 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 661 return -EINVAL; 662 663 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 664 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 665 AF_INET, prefixlen, l3index, flags, 666 cmd.tcpm_key, cmd.tcpm_keylen); 667 668 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 669 AF_INET6, prefixlen, l3index, flags, 670 cmd.tcpm_key, cmd.tcpm_keylen); 671 } 672 673 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp, 674 const struct in6_addr *daddr, 675 const struct in6_addr *saddr, 676 const struct tcphdr *th, int nbytes) 677 { 678 struct tcp6_pseudohdr *bp; 679 struct scatterlist sg; 680 struct tcphdr *_th; 681 682 bp = hp->scratch; 683 /* 1. TCP pseudo-header (RFC2460) */ 684 bp->saddr = *saddr; 685 bp->daddr = *daddr; 686 bp->protocol = cpu_to_be32(IPPROTO_TCP); 687 bp->len = cpu_to_be32(nbytes); 688 689 _th = (struct tcphdr *)(bp + 1); 690 memcpy(_th, th, sizeof(*th)); 691 _th->check = 0; 692 693 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); 694 ahash_request_set_crypt(hp->md5_req, &sg, NULL, 695 sizeof(*bp) + sizeof(*th)); 696 return crypto_ahash_update(hp->md5_req); 697 } 698 699 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 700 const struct in6_addr *daddr, struct in6_addr *saddr, 701 const struct tcphdr *th) 702 { 703 struct tcp_md5sig_pool *hp; 704 struct ahash_request *req; 705 706 hp = tcp_get_md5sig_pool(); 707 if (!hp) 708 goto clear_hash_noput; 709 req = hp->md5_req; 710 711 if (crypto_ahash_init(req)) 712 goto clear_hash; 713 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) 714 goto clear_hash; 715 if (tcp_md5_hash_key(hp, key)) 716 goto clear_hash; 717 ahash_request_set_crypt(req, NULL, md5_hash, 0); 718 if (crypto_ahash_final(req)) 719 goto clear_hash; 720 721 tcp_put_md5sig_pool(); 722 return 0; 723 724 clear_hash: 725 tcp_put_md5sig_pool(); 726 clear_hash_noput: 727 memset(md5_hash, 0, 16); 728 return 1; 729 } 730 731 static int tcp_v6_md5_hash_skb(char *md5_hash, 732 const struct tcp_md5sig_key *key, 733 const struct sock *sk, 734 const struct sk_buff *skb) 735 { 736 const struct in6_addr *saddr, *daddr; 737 struct tcp_md5sig_pool *hp; 738 struct ahash_request *req; 739 const struct tcphdr *th = tcp_hdr(skb); 740 741 if (sk) { /* valid for establish/request sockets */ 742 saddr = &sk->sk_v6_rcv_saddr; 743 daddr = &sk->sk_v6_daddr; 744 } else { 745 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 746 saddr = &ip6h->saddr; 747 daddr = &ip6h->daddr; 748 } 749 750 hp = tcp_get_md5sig_pool(); 751 if (!hp) 752 goto clear_hash_noput; 753 req = hp->md5_req; 754 755 if (crypto_ahash_init(req)) 756 goto clear_hash; 757 758 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len)) 759 goto clear_hash; 760 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 761 goto clear_hash; 762 if (tcp_md5_hash_key(hp, key)) 763 goto clear_hash; 764 ahash_request_set_crypt(req, NULL, md5_hash, 0); 765 if (crypto_ahash_final(req)) 766 goto clear_hash; 767 768 tcp_put_md5sig_pool(); 769 return 0; 770 771 clear_hash: 772 tcp_put_md5sig_pool(); 773 clear_hash_noput: 774 memset(md5_hash, 0, 16); 775 return 1; 776 } 777 778 #endif 779 780 static void tcp_v6_init_req(struct request_sock *req, 781 const struct sock *sk_listener, 782 struct sk_buff *skb) 783 { 784 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 785 struct inet_request_sock *ireq = inet_rsk(req); 786 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener); 787 788 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; 789 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; 790 791 /* So that link locals have meaning */ 792 if ((!sk_listener->sk_bound_dev_if || l3_slave) && 793 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) 794 ireq->ir_iif = tcp_v6_iif(skb); 795 796 if (!TCP_SKB_CB(skb)->tcp_tw_isn && 797 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) || 798 np->rxopt.bits.rxinfo || 799 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || 800 np->rxopt.bits.rxohlim || inet6_test_bit(REPFLOW, sk_listener))) { 801 refcount_inc(&skb->users); 802 ireq->pktopts = skb; 803 } 804 } 805 806 static struct dst_entry *tcp_v6_route_req(const struct sock *sk, 807 struct sk_buff *skb, 808 struct flowi *fl, 809 struct request_sock *req) 810 { 811 tcp_v6_init_req(req, sk, skb); 812 813 if (security_inet_conn_request(sk, skb, req)) 814 return NULL; 815 816 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP); 817 } 818 819 struct request_sock_ops tcp6_request_sock_ops __read_mostly = { 820 .family = AF_INET6, 821 .obj_size = sizeof(struct tcp6_request_sock), 822 .rtx_syn_ack = tcp_rtx_synack, 823 .send_ack = tcp_v6_reqsk_send_ack, 824 .destructor = tcp_v6_reqsk_destructor, 825 .send_reset = tcp_v6_send_reset, 826 .syn_ack_timeout = tcp_syn_ack_timeout, 827 }; 828 829 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { 830 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 831 sizeof(struct ipv6hdr), 832 #ifdef CONFIG_TCP_MD5SIG 833 .req_md5_lookup = tcp_v6_md5_lookup, 834 .calc_md5_hash = tcp_v6_md5_hash_skb, 835 #endif 836 #ifdef CONFIG_SYN_COOKIES 837 .cookie_init_seq = cookie_v6_init_sequence, 838 #endif 839 .route_req = tcp_v6_route_req, 840 .init_seq = tcp_v6_init_seq, 841 .init_ts_off = tcp_v6_init_ts_off, 842 .send_synack = tcp_v6_send_synack, 843 }; 844 845 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, 846 u32 ack, u32 win, u32 tsval, u32 tsecr, 847 int oif, struct tcp_md5sig_key *key, int rst, 848 u8 tclass, __be32 label, u32 priority, u32 txhash) 849 { 850 const struct tcphdr *th = tcp_hdr(skb); 851 struct tcphdr *t1; 852 struct sk_buff *buff; 853 struct flowi6 fl6; 854 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 855 struct sock *ctl_sk = net->ipv6.tcp_sk; 856 unsigned int tot_len = sizeof(struct tcphdr); 857 __be32 mrst = 0, *topt; 858 struct dst_entry *dst; 859 __u32 mark = 0; 860 861 if (tsecr) 862 tot_len += TCPOLEN_TSTAMP_ALIGNED; 863 #ifdef CONFIG_TCP_MD5SIG 864 if (key) 865 tot_len += TCPOLEN_MD5SIG_ALIGNED; 866 #endif 867 868 #ifdef CONFIG_MPTCP 869 if (rst && !key) { 870 mrst = mptcp_reset_option(skb); 871 872 if (mrst) 873 tot_len += sizeof(__be32); 874 } 875 #endif 876 877 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC); 878 if (!buff) 879 return; 880 881 skb_reserve(buff, MAX_TCP_HEADER); 882 883 t1 = skb_push(buff, tot_len); 884 skb_reset_transport_header(buff); 885 886 /* Swap the send and the receive. */ 887 memset(t1, 0, sizeof(*t1)); 888 t1->dest = th->source; 889 t1->source = th->dest; 890 t1->doff = tot_len / 4; 891 t1->seq = htonl(seq); 892 t1->ack_seq = htonl(ack); 893 t1->ack = !rst || !th->ack; 894 t1->rst = rst; 895 t1->window = htons(win); 896 897 topt = (__be32 *)(t1 + 1); 898 899 if (tsecr) { 900 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 901 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 902 *topt++ = htonl(tsval); 903 *topt++ = htonl(tsecr); 904 } 905 906 if (mrst) 907 *topt++ = mrst; 908 909 #ifdef CONFIG_TCP_MD5SIG 910 if (key) { 911 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 912 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); 913 tcp_v6_md5_hash_hdr((__u8 *)topt, key, 914 &ipv6_hdr(skb)->saddr, 915 &ipv6_hdr(skb)->daddr, t1); 916 } 917 #endif 918 919 memset(&fl6, 0, sizeof(fl6)); 920 fl6.daddr = ipv6_hdr(skb)->saddr; 921 fl6.saddr = ipv6_hdr(skb)->daddr; 922 fl6.flowlabel = label; 923 924 buff->ip_summed = CHECKSUM_PARTIAL; 925 926 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr); 927 928 fl6.flowi6_proto = IPPROTO_TCP; 929 if (rt6_need_strict(&fl6.daddr) && !oif) 930 fl6.flowi6_oif = tcp_v6_iif(skb); 931 else { 932 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 933 oif = skb->skb_iif; 934 935 fl6.flowi6_oif = oif; 936 } 937 938 if (sk) { 939 if (sk->sk_state == TCP_TIME_WAIT) 940 mark = inet_twsk(sk)->tw_mark; 941 else 942 mark = READ_ONCE(sk->sk_mark); 943 skb_set_delivery_time(buff, tcp_transmit_time(sk), true); 944 } 945 if (txhash) { 946 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */ 947 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4); 948 } 949 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark; 950 fl6.fl6_dport = t1->dest; 951 fl6.fl6_sport = t1->source; 952 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); 953 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6)); 954 955 /* Pass a socket to ip6_dst_lookup either it is for RST 956 * Underlying function will use this to retrieve the network 957 * namespace 958 */ 959 if (sk && sk->sk_state != TCP_TIME_WAIT) 960 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/ 961 else 962 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL); 963 if (!IS_ERR(dst)) { 964 skb_dst_set(buff, dst); 965 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, 966 tclass & ~INET_ECN_MASK, priority); 967 TCP_INC_STATS(net, TCP_MIB_OUTSEGS); 968 if (rst) 969 TCP_INC_STATS(net, TCP_MIB_OUTRSTS); 970 return; 971 } 972 973 kfree_skb(buff); 974 } 975 976 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) 977 { 978 const struct tcphdr *th = tcp_hdr(skb); 979 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 980 u32 seq = 0, ack_seq = 0; 981 struct tcp_md5sig_key *key = NULL; 982 #ifdef CONFIG_TCP_MD5SIG 983 const __u8 *hash_location = NULL; 984 unsigned char newhash[16]; 985 int genhash; 986 struct sock *sk1 = NULL; 987 #endif 988 __be32 label = 0; 989 u32 priority = 0; 990 struct net *net; 991 u32 txhash = 0; 992 int oif = 0; 993 994 if (th->rst) 995 return; 996 997 /* If sk not NULL, it means we did a successful lookup and incoming 998 * route had to be correct. prequeue might have dropped our dst. 999 */ 1000 if (!sk && !ipv6_unicast_destination(skb)) 1001 return; 1002 1003 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 1004 #ifdef CONFIG_TCP_MD5SIG 1005 rcu_read_lock(); 1006 hash_location = tcp_parse_md5sig_option(th); 1007 if (sk && sk_fullsock(sk)) { 1008 int l3index; 1009 1010 /* sdif set, means packet ingressed via a device 1011 * in an L3 domain and inet_iif is set to it. 1012 */ 1013 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1014 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index); 1015 } else if (hash_location) { 1016 int dif = tcp_v6_iif_l3_slave(skb); 1017 int sdif = tcp_v6_sdif(skb); 1018 int l3index; 1019 1020 /* 1021 * active side is lost. Try to find listening socket through 1022 * source port, and then find md5 key through listening socket. 1023 * we are not loose security here: 1024 * Incoming packet is checked with md5 hash with finding key, 1025 * no RST generated if md5 hash doesn't match. 1026 */ 1027 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1028 NULL, 0, &ipv6h->saddr, th->source, 1029 &ipv6h->daddr, ntohs(th->source), 1030 dif, sdif); 1031 if (!sk1) 1032 goto out; 1033 1034 /* sdif set, means packet ingressed via a device 1035 * in an L3 domain and dif is set to it. 1036 */ 1037 l3index = tcp_v6_sdif(skb) ? dif : 0; 1038 1039 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index); 1040 if (!key) 1041 goto out; 1042 1043 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb); 1044 if (genhash || memcmp(hash_location, newhash, 16) != 0) 1045 goto out; 1046 } 1047 #endif 1048 1049 if (th->ack) 1050 seq = ntohl(th->ack_seq); 1051 else 1052 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - 1053 (th->doff << 2); 1054 1055 if (sk) { 1056 oif = sk->sk_bound_dev_if; 1057 if (sk_fullsock(sk)) { 1058 trace_tcp_send_reset(sk, skb); 1059 if (inet6_test_bit(REPFLOW, sk)) 1060 label = ip6_flowlabel(ipv6h); 1061 priority = READ_ONCE(sk->sk_priority); 1062 txhash = sk->sk_txhash; 1063 } 1064 if (sk->sk_state == TCP_TIME_WAIT) { 1065 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel); 1066 priority = inet_twsk(sk)->tw_priority; 1067 txhash = inet_twsk(sk)->tw_txhash; 1068 } 1069 } else { 1070 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET) 1071 label = ip6_flowlabel(ipv6h); 1072 } 1073 1074 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 1075 ipv6_get_dsfield(ipv6h), label, priority, txhash); 1076 1077 #ifdef CONFIG_TCP_MD5SIG 1078 out: 1079 rcu_read_unlock(); 1080 #endif 1081 } 1082 1083 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, 1084 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, 1085 struct tcp_md5sig_key *key, u8 tclass, 1086 __be32 label, u32 priority, u32 txhash) 1087 { 1088 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0, 1089 tclass, label, priority, txhash); 1090 } 1091 1092 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) 1093 { 1094 struct inet_timewait_sock *tw = inet_twsk(sk); 1095 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 1096 1097 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 1098 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 1099 tcp_time_stamp_raw() + tcptw->tw_ts_offset, 1100 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw), 1101 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority, 1102 tw->tw_txhash); 1103 1104 inet_twsk_put(tw); 1105 } 1106 1107 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 1108 struct request_sock *req) 1109 { 1110 int l3index; 1111 1112 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1113 1114 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV 1115 * sk->sk_state == TCP_SYN_RECV -> for Fast Open. 1116 */ 1117 /* RFC 7323 2.3 1118 * The window field (SEG.WND) of every outgoing segment, with the 1119 * exception of <SYN> segments, MUST be right-shifted by 1120 * Rcv.Wind.Shift bits: 1121 */ 1122 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ? 1123 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, 1124 tcp_rsk(req)->rcv_nxt, 1125 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, 1126 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off, 1127 READ_ONCE(req->ts_recent), sk->sk_bound_dev_if, 1128 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index), 1129 ipv6_get_dsfield(ipv6_hdr(skb)), 0, 1130 READ_ONCE(sk->sk_priority), 1131 READ_ONCE(tcp_rsk(req)->txhash)); 1132 } 1133 1134 1135 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) 1136 { 1137 #ifdef CONFIG_SYN_COOKIES 1138 const struct tcphdr *th = tcp_hdr(skb); 1139 1140 if (!th->syn) 1141 sk = cookie_v6_check(sk, skb); 1142 #endif 1143 return sk; 1144 } 1145 1146 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 1147 struct tcphdr *th, u32 *cookie) 1148 { 1149 u16 mss = 0; 1150 #ifdef CONFIG_SYN_COOKIES 1151 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops, 1152 &tcp_request_sock_ipv6_ops, sk, th); 1153 if (mss) { 1154 *cookie = __cookie_v6_init_sequence(iph, th, &mss); 1155 tcp_synq_overflow(sk); 1156 } 1157 #endif 1158 return mss; 1159 } 1160 1161 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 1162 { 1163 if (skb->protocol == htons(ETH_P_IP)) 1164 return tcp_v4_conn_request(sk, skb); 1165 1166 if (!ipv6_unicast_destination(skb)) 1167 goto drop; 1168 1169 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) { 1170 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); 1171 return 0; 1172 } 1173 1174 return tcp_conn_request(&tcp6_request_sock_ops, 1175 &tcp_request_sock_ipv6_ops, sk, skb); 1176 1177 drop: 1178 tcp_listendrop(sk); 1179 return 0; /* don't send reset */ 1180 } 1181 1182 static void tcp_v6_restore_cb(struct sk_buff *skb) 1183 { 1184 /* We need to move header back to the beginning if xfrm6_policy_check() 1185 * and tcp_v6_fill_cb() are going to be called again. 1186 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there. 1187 */ 1188 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, 1189 sizeof(struct inet6_skb_parm)); 1190 } 1191 1192 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 1193 struct request_sock *req, 1194 struct dst_entry *dst, 1195 struct request_sock *req_unhash, 1196 bool *own_req) 1197 { 1198 struct inet_request_sock *ireq; 1199 struct ipv6_pinfo *newnp; 1200 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1201 struct ipv6_txoptions *opt; 1202 struct inet_sock *newinet; 1203 bool found_dup_sk = false; 1204 struct tcp_sock *newtp; 1205 struct sock *newsk; 1206 #ifdef CONFIG_TCP_MD5SIG 1207 struct tcp_md5sig_key *key; 1208 int l3index; 1209 #endif 1210 struct flowi6 fl6; 1211 1212 if (skb->protocol == htons(ETH_P_IP)) { 1213 /* 1214 * v6 mapped 1215 */ 1216 1217 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, 1218 req_unhash, own_req); 1219 1220 if (!newsk) 1221 return NULL; 1222 1223 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1224 1225 newnp = tcp_inet6_sk(newsk); 1226 newtp = tcp_sk(newsk); 1227 1228 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1229 1230 newnp->saddr = newsk->sk_v6_rcv_saddr; 1231 1232 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; 1233 if (sk_is_mptcp(newsk)) 1234 mptcpv6_handle_mapped(newsk, true); 1235 newsk->sk_backlog_rcv = tcp_v4_do_rcv; 1236 #ifdef CONFIG_TCP_MD5SIG 1237 newtp->af_specific = &tcp_sock_ipv6_mapped_specific; 1238 #endif 1239 1240 newnp->ipv6_mc_list = NULL; 1241 newnp->ipv6_ac_list = NULL; 1242 newnp->ipv6_fl_list = NULL; 1243 newnp->pktoptions = NULL; 1244 newnp->opt = NULL; 1245 newnp->mcast_oif = inet_iif(skb); 1246 newnp->mcast_hops = ip_hdr(skb)->ttl; 1247 newnp->rcv_flowinfo = 0; 1248 if (inet6_test_bit(REPFLOW, sk)) 1249 newnp->flow_label = 0; 1250 1251 /* 1252 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 1253 * here, tcp_create_openreq_child now does this for us, see the comment in 1254 * that function for the gory details. -acme 1255 */ 1256 1257 /* It is tricky place. Until this moment IPv4 tcp 1258 worked with IPv6 icsk.icsk_af_ops. 1259 Sync it now. 1260 */ 1261 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 1262 1263 return newsk; 1264 } 1265 1266 ireq = inet_rsk(req); 1267 1268 if (sk_acceptq_is_full(sk)) 1269 goto out_overflow; 1270 1271 if (!dst) { 1272 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP); 1273 if (!dst) 1274 goto out; 1275 } 1276 1277 newsk = tcp_create_openreq_child(sk, req, skb); 1278 if (!newsk) 1279 goto out_nonewsk; 1280 1281 /* 1282 * No need to charge this sock to the relevant IPv6 refcnt debug socks 1283 * count here, tcp_create_openreq_child now does this for us, see the 1284 * comment in that function for the gory details. -acme 1285 */ 1286 1287 newsk->sk_gso_type = SKB_GSO_TCPV6; 1288 ip6_dst_store(newsk, dst, NULL, NULL); 1289 inet6_sk_rx_dst_set(newsk, skb); 1290 1291 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1292 1293 newtp = tcp_sk(newsk); 1294 newinet = inet_sk(newsk); 1295 newnp = tcp_inet6_sk(newsk); 1296 1297 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1298 1299 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr; 1300 newnp->saddr = ireq->ir_v6_loc_addr; 1301 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr; 1302 newsk->sk_bound_dev_if = ireq->ir_iif; 1303 1304 /* Now IPv6 options... 1305 1306 First: no IPv4 options. 1307 */ 1308 newinet->inet_opt = NULL; 1309 newnp->ipv6_mc_list = NULL; 1310 newnp->ipv6_ac_list = NULL; 1311 newnp->ipv6_fl_list = NULL; 1312 1313 /* Clone RX bits */ 1314 newnp->rxopt.all = np->rxopt.all; 1315 1316 newnp->pktoptions = NULL; 1317 newnp->opt = NULL; 1318 newnp->mcast_oif = tcp_v6_iif(skb); 1319 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1320 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); 1321 if (inet6_test_bit(REPFLOW, sk)) 1322 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); 1323 1324 /* Set ToS of the new socket based upon the value of incoming SYN. 1325 * ECT bits are set later in tcp_init_transfer(). 1326 */ 1327 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)) 1328 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK; 1329 1330 /* Clone native IPv6 options from listening socket (if any) 1331 1332 Yes, keeping reference count would be much more clever, 1333 but we make one more one thing there: reattach optmem 1334 to newsk. 1335 */ 1336 opt = ireq->ipv6_opt; 1337 if (!opt) 1338 opt = rcu_dereference(np->opt); 1339 if (opt) { 1340 opt = ipv6_dup_options(newsk, opt); 1341 RCU_INIT_POINTER(newnp->opt, opt); 1342 } 1343 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1344 if (opt) 1345 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen + 1346 opt->opt_flen; 1347 1348 tcp_ca_openreq_child(newsk, dst); 1349 1350 tcp_sync_mss(newsk, dst_mtu(dst)); 1351 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst)); 1352 1353 tcp_initialize_rcv_mss(newsk); 1354 1355 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 1356 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 1357 1358 #ifdef CONFIG_TCP_MD5SIG 1359 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif); 1360 1361 /* Copy over the MD5 key from the original socket */ 1362 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index); 1363 if (key) { 1364 const union tcp_md5_addr *addr; 1365 1366 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr; 1367 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) { 1368 inet_csk_prepare_forced_close(newsk); 1369 tcp_done(newsk); 1370 goto out; 1371 } 1372 } 1373 #endif 1374 1375 if (__inet_inherit_port(sk, newsk) < 0) { 1376 inet_csk_prepare_forced_close(newsk); 1377 tcp_done(newsk); 1378 goto out; 1379 } 1380 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), 1381 &found_dup_sk); 1382 if (*own_req) { 1383 tcp_move_syn(newtp, req); 1384 1385 /* Clone pktoptions received with SYN, if we own the req */ 1386 if (ireq->pktopts) { 1387 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk); 1388 consume_skb(ireq->pktopts); 1389 ireq->pktopts = NULL; 1390 if (newnp->pktoptions) 1391 tcp_v6_restore_cb(newnp->pktoptions); 1392 } 1393 } else { 1394 if (!req_unhash && found_dup_sk) { 1395 /* This code path should only be executed in the 1396 * syncookie case only 1397 */ 1398 bh_unlock_sock(newsk); 1399 sock_put(newsk); 1400 newsk = NULL; 1401 } 1402 } 1403 1404 return newsk; 1405 1406 out_overflow: 1407 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1408 out_nonewsk: 1409 dst_release(dst); 1410 out: 1411 tcp_listendrop(sk); 1412 return NULL; 1413 } 1414 1415 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 1416 u32)); 1417 /* The socket must have it's spinlock held when we get 1418 * here, unless it is a TCP_LISTEN socket. 1419 * 1420 * We have a potential double-lock case here, so even when 1421 * doing backlog processing we use the BH locking scheme. 1422 * This is because we cannot sleep with the original spinlock 1423 * held. 1424 */ 1425 INDIRECT_CALLABLE_SCOPE 1426 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 1427 { 1428 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1429 struct sk_buff *opt_skb = NULL; 1430 enum skb_drop_reason reason; 1431 struct tcp_sock *tp; 1432 1433 /* Imagine: socket is IPv6. IPv4 packet arrives, 1434 goes to IPv4 receive handler and backlogged. 1435 From backlog it always goes here. Kerboom... 1436 Fortunately, tcp_rcv_established and rcv_established 1437 handle them correctly, but it is not case with 1438 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK 1439 */ 1440 1441 if (skb->protocol == htons(ETH_P_IP)) 1442 return tcp_v4_do_rcv(sk, skb); 1443 1444 /* 1445 * socket locking is here for SMP purposes as backlog rcv 1446 * is currently called with bh processing disabled. 1447 */ 1448 1449 /* Do Stevens' IPV6_PKTOPTIONS. 1450 1451 Yes, guys, it is the only place in our code, where we 1452 may make it not affecting IPv4. 1453 The rest of code is protocol independent, 1454 and I do not like idea to uglify IPv4. 1455 1456 Actually, all the idea behind IPV6_PKTOPTIONS 1457 looks not very well thought. For now we latch 1458 options, received in the last packet, enqueued 1459 by tcp. Feel free to propose better solution. 1460 --ANK (980728) 1461 */ 1462 if (np->rxopt.all) 1463 opt_skb = skb_clone_and_charge_r(skb, sk); 1464 1465 reason = SKB_DROP_REASON_NOT_SPECIFIED; 1466 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1467 struct dst_entry *dst; 1468 1469 dst = rcu_dereference_protected(sk->sk_rx_dst, 1470 lockdep_sock_is_held(sk)); 1471 1472 sock_rps_save_rxhash(sk, skb); 1473 sk_mark_napi_id(sk, skb); 1474 if (dst) { 1475 if (sk->sk_rx_dst_ifindex != skb->skb_iif || 1476 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check, 1477 dst, sk->sk_rx_dst_cookie) == NULL) { 1478 RCU_INIT_POINTER(sk->sk_rx_dst, NULL); 1479 dst_release(dst); 1480 } 1481 } 1482 1483 tcp_rcv_established(sk, skb); 1484 if (opt_skb) 1485 goto ipv6_pktoptions; 1486 return 0; 1487 } 1488 1489 if (tcp_checksum_complete(skb)) 1490 goto csum_err; 1491 1492 if (sk->sk_state == TCP_LISTEN) { 1493 struct sock *nsk = tcp_v6_cookie_check(sk, skb); 1494 1495 if (!nsk) 1496 goto discard; 1497 1498 if (nsk != sk) { 1499 if (tcp_child_process(sk, nsk, skb)) 1500 goto reset; 1501 if (opt_skb) 1502 __kfree_skb(opt_skb); 1503 return 0; 1504 } 1505 } else 1506 sock_rps_save_rxhash(sk, skb); 1507 1508 if (tcp_rcv_state_process(sk, skb)) 1509 goto reset; 1510 if (opt_skb) 1511 goto ipv6_pktoptions; 1512 return 0; 1513 1514 reset: 1515 tcp_v6_send_reset(sk, skb); 1516 discard: 1517 if (opt_skb) 1518 __kfree_skb(opt_skb); 1519 kfree_skb_reason(skb, reason); 1520 return 0; 1521 csum_err: 1522 reason = SKB_DROP_REASON_TCP_CSUM; 1523 trace_tcp_bad_csum(skb); 1524 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); 1525 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); 1526 goto discard; 1527 1528 1529 ipv6_pktoptions: 1530 /* Do you ask, what is it? 1531 1532 1. skb was enqueued by tcp. 1533 2. skb is added to tail of read queue, rather than out of order. 1534 3. socket is not in passive state. 1535 4. Finally, it really contains options, which user wants to receive. 1536 */ 1537 tp = tcp_sk(sk); 1538 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && 1539 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { 1540 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) 1541 np->mcast_oif = tcp_v6_iif(opt_skb); 1542 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) 1543 WRITE_ONCE(np->mcast_hops, 1544 ipv6_hdr(opt_skb)->hop_limit); 1545 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) 1546 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb)); 1547 if (inet6_test_bit(REPFLOW, sk)) 1548 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb)); 1549 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) { 1550 tcp_v6_restore_cb(opt_skb); 1551 opt_skb = xchg(&np->pktoptions, opt_skb); 1552 } else { 1553 __kfree_skb(opt_skb); 1554 opt_skb = xchg(&np->pktoptions, NULL); 1555 } 1556 } 1557 1558 consume_skb(opt_skb); 1559 return 0; 1560 } 1561 1562 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, 1563 const struct tcphdr *th) 1564 { 1565 /* This is tricky: we move IP6CB at its correct location into 1566 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because 1567 * _decode_session6() uses IP6CB(). 1568 * barrier() makes sure compiler won't play aliasing games. 1569 */ 1570 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), 1571 sizeof(struct inet6_skb_parm)); 1572 barrier(); 1573 1574 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1575 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1576 skb->len - th->doff*4); 1577 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1578 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1579 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1580 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr); 1581 TCP_SKB_CB(skb)->sacked = 0; 1582 TCP_SKB_CB(skb)->has_rxtstamp = 1583 skb->tstamp || skb_hwtstamps(skb)->hwtstamp; 1584 } 1585 1586 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) 1587 { 1588 enum skb_drop_reason drop_reason; 1589 int sdif = inet6_sdif(skb); 1590 int dif = inet6_iif(skb); 1591 const struct tcphdr *th; 1592 const struct ipv6hdr *hdr; 1593 bool refcounted; 1594 struct sock *sk; 1595 int ret; 1596 struct net *net = dev_net(skb->dev); 1597 1598 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 1599 if (skb->pkt_type != PACKET_HOST) 1600 goto discard_it; 1601 1602 /* 1603 * Count it even if it's bad. 1604 */ 1605 __TCP_INC_STATS(net, TCP_MIB_INSEGS); 1606 1607 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1608 goto discard_it; 1609 1610 th = (const struct tcphdr *)skb->data; 1611 1612 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) { 1613 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1614 goto bad_packet; 1615 } 1616 if (!pskb_may_pull(skb, th->doff*4)) 1617 goto discard_it; 1618 1619 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) 1620 goto csum_error; 1621 1622 th = (const struct tcphdr *)skb->data; 1623 hdr = ipv6_hdr(skb); 1624 1625 lookup: 1626 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th), 1627 th->source, th->dest, inet6_iif(skb), sdif, 1628 &refcounted); 1629 if (!sk) 1630 goto no_tcp_socket; 1631 1632 process: 1633 if (sk->sk_state == TCP_TIME_WAIT) 1634 goto do_time_wait; 1635 1636 if (sk->sk_state == TCP_NEW_SYN_RECV) { 1637 struct request_sock *req = inet_reqsk(sk); 1638 bool req_stolen = false; 1639 struct sock *nsk; 1640 1641 sk = req->rsk_listener; 1642 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 1643 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1644 else 1645 drop_reason = tcp_inbound_md5_hash(sk, skb, 1646 &hdr->saddr, &hdr->daddr, 1647 AF_INET6, dif, sdif); 1648 if (drop_reason) { 1649 sk_drops_add(sk, skb); 1650 reqsk_put(req); 1651 goto discard_it; 1652 } 1653 if (tcp_checksum_complete(skb)) { 1654 reqsk_put(req); 1655 goto csum_error; 1656 } 1657 if (unlikely(sk->sk_state != TCP_LISTEN)) { 1658 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb); 1659 if (!nsk) { 1660 inet_csk_reqsk_queue_drop_and_put(sk, req); 1661 goto lookup; 1662 } 1663 sk = nsk; 1664 /* reuseport_migrate_sock() has already held one sk_refcnt 1665 * before returning. 1666 */ 1667 } else { 1668 sock_hold(sk); 1669 } 1670 refcounted = true; 1671 nsk = NULL; 1672 if (!tcp_filter(sk, skb)) { 1673 th = (const struct tcphdr *)skb->data; 1674 hdr = ipv6_hdr(skb); 1675 tcp_v6_fill_cb(skb, hdr, th); 1676 nsk = tcp_check_req(sk, skb, req, false, &req_stolen); 1677 } else { 1678 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1679 } 1680 if (!nsk) { 1681 reqsk_put(req); 1682 if (req_stolen) { 1683 /* Another cpu got exclusive access to req 1684 * and created a full blown socket. 1685 * Try to feed this packet to this socket 1686 * instead of discarding it. 1687 */ 1688 tcp_v6_restore_cb(skb); 1689 sock_put(sk); 1690 goto lookup; 1691 } 1692 goto discard_and_relse; 1693 } 1694 nf_reset_ct(skb); 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 nf_reset_ct(skb); 1727 1728 if (tcp_filter(sk, skb)) { 1729 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1730 goto discard_and_relse; 1731 } 1732 th = (const struct tcphdr *)skb->data; 1733 hdr = ipv6_hdr(skb); 1734 tcp_v6_fill_cb(skb, hdr, th); 1735 1736 skb->dev = NULL; 1737 1738 if (sk->sk_state == TCP_LISTEN) { 1739 ret = tcp_v6_do_rcv(sk, skb); 1740 goto put_and_return; 1741 } 1742 1743 sk_incoming_cpu_update(sk); 1744 1745 bh_lock_sock_nested(sk); 1746 tcp_segs_in(tcp_sk(sk), skb); 1747 ret = 0; 1748 if (!sock_owned_by_user(sk)) { 1749 ret = tcp_v6_do_rcv(sk, skb); 1750 } else { 1751 if (tcp_add_backlog(sk, skb, &drop_reason)) 1752 goto discard_and_relse; 1753 } 1754 bh_unlock_sock(sk); 1755 put_and_return: 1756 if (refcounted) 1757 sock_put(sk); 1758 return ret ? -1 : 0; 1759 1760 no_tcp_socket: 1761 drop_reason = SKB_DROP_REASON_NO_SOCKET; 1762 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1763 goto discard_it; 1764 1765 tcp_v6_fill_cb(skb, hdr, th); 1766 1767 if (tcp_checksum_complete(skb)) { 1768 csum_error: 1769 drop_reason = SKB_DROP_REASON_TCP_CSUM; 1770 trace_tcp_bad_csum(skb); 1771 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); 1772 bad_packet: 1773 __TCP_INC_STATS(net, TCP_MIB_INERRS); 1774 } else { 1775 tcp_v6_send_reset(NULL, skb); 1776 } 1777 1778 discard_it: 1779 SKB_DR_OR(drop_reason, NOT_SPECIFIED); 1780 kfree_skb_reason(skb, drop_reason); 1781 return 0; 1782 1783 discard_and_relse: 1784 sk_drops_add(sk, skb); 1785 if (refcounted) 1786 sock_put(sk); 1787 goto discard_it; 1788 1789 do_time_wait: 1790 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1791 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1792 inet_twsk_put(inet_twsk(sk)); 1793 goto discard_it; 1794 } 1795 1796 tcp_v6_fill_cb(skb, hdr, th); 1797 1798 if (tcp_checksum_complete(skb)) { 1799 inet_twsk_put(inet_twsk(sk)); 1800 goto csum_error; 1801 } 1802 1803 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1804 case TCP_TW_SYN: 1805 { 1806 struct sock *sk2; 1807 1808 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1809 skb, __tcp_hdrlen(th), 1810 &ipv6_hdr(skb)->saddr, th->source, 1811 &ipv6_hdr(skb)->daddr, 1812 ntohs(th->dest), 1813 tcp_v6_iif_l3_slave(skb), 1814 sdif); 1815 if (sk2) { 1816 struct inet_timewait_sock *tw = inet_twsk(sk); 1817 inet_twsk_deschedule_put(tw); 1818 sk = sk2; 1819 tcp_v6_restore_cb(skb); 1820 refcounted = false; 1821 goto process; 1822 } 1823 } 1824 /* to ACK */ 1825 fallthrough; 1826 case TCP_TW_ACK: 1827 tcp_v6_timewait_ack(sk, skb); 1828 break; 1829 case TCP_TW_RST: 1830 tcp_v6_send_reset(sk, skb); 1831 inet_twsk_deschedule_put(inet_twsk(sk)); 1832 goto discard_it; 1833 case TCP_TW_SUCCESS: 1834 ; 1835 } 1836 goto discard_it; 1837 } 1838 1839 void tcp_v6_early_demux(struct sk_buff *skb) 1840 { 1841 struct net *net = dev_net(skb->dev); 1842 const struct ipv6hdr *hdr; 1843 const struct tcphdr *th; 1844 struct sock *sk; 1845 1846 if (skb->pkt_type != PACKET_HOST) 1847 return; 1848 1849 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1850 return; 1851 1852 hdr = ipv6_hdr(skb); 1853 th = tcp_hdr(skb); 1854 1855 if (th->doff < sizeof(struct tcphdr) / 4) 1856 return; 1857 1858 /* Note : We use inet6_iif() here, not tcp_v6_iif() */ 1859 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 1860 &hdr->saddr, th->source, 1861 &hdr->daddr, ntohs(th->dest), 1862 inet6_iif(skb), inet6_sdif(skb)); 1863 if (sk) { 1864 skb->sk = sk; 1865 skb->destructor = sock_edemux; 1866 if (sk_fullsock(sk)) { 1867 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst); 1868 1869 if (dst) 1870 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1871 if (dst && 1872 sk->sk_rx_dst_ifindex == skb->skb_iif) 1873 skb_dst_set_noref(skb, dst); 1874 } 1875 } 1876 } 1877 1878 static struct timewait_sock_ops tcp6_timewait_sock_ops = { 1879 .twsk_obj_size = sizeof(struct tcp6_timewait_sock), 1880 .twsk_unique = tcp_twsk_unique, 1881 .twsk_destructor = tcp_twsk_destructor, 1882 }; 1883 1884 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb) 1885 { 1886 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr); 1887 } 1888 1889 const struct inet_connection_sock_af_ops ipv6_specific = { 1890 .queue_xmit = inet6_csk_xmit, 1891 .send_check = tcp_v6_send_check, 1892 .rebuild_header = inet6_sk_rebuild_header, 1893 .sk_rx_dst_set = inet6_sk_rx_dst_set, 1894 .conn_request = tcp_v6_conn_request, 1895 .syn_recv_sock = tcp_v6_syn_recv_sock, 1896 .net_header_len = sizeof(struct ipv6hdr), 1897 .net_frag_header_len = sizeof(struct frag_hdr), 1898 .setsockopt = ipv6_setsockopt, 1899 .getsockopt = ipv6_getsockopt, 1900 .addr2sockaddr = inet6_csk_addr2sockaddr, 1901 .sockaddr_len = sizeof(struct sockaddr_in6), 1902 .mtu_reduced = tcp_v6_mtu_reduced, 1903 }; 1904 1905 #ifdef CONFIG_TCP_MD5SIG 1906 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { 1907 .md5_lookup = tcp_v6_md5_lookup, 1908 .calc_md5_hash = tcp_v6_md5_hash_skb, 1909 .md5_parse = tcp_v6_parse_md5_keys, 1910 }; 1911 #endif 1912 1913 /* 1914 * TCP over IPv4 via INET6 API 1915 */ 1916 static const struct inet_connection_sock_af_ops ipv6_mapped = { 1917 .queue_xmit = ip_queue_xmit, 1918 .send_check = tcp_v4_send_check, 1919 .rebuild_header = inet_sk_rebuild_header, 1920 .sk_rx_dst_set = inet_sk_rx_dst_set, 1921 .conn_request = tcp_v6_conn_request, 1922 .syn_recv_sock = tcp_v6_syn_recv_sock, 1923 .net_header_len = sizeof(struct iphdr), 1924 .setsockopt = ipv6_setsockopt, 1925 .getsockopt = ipv6_getsockopt, 1926 .addr2sockaddr = inet6_csk_addr2sockaddr, 1927 .sockaddr_len = sizeof(struct sockaddr_in6), 1928 .mtu_reduced = tcp_v4_mtu_reduced, 1929 }; 1930 1931 #ifdef CONFIG_TCP_MD5SIG 1932 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { 1933 .md5_lookup = tcp_v4_md5_lookup, 1934 .calc_md5_hash = tcp_v4_md5_hash_skb, 1935 .md5_parse = tcp_v6_parse_md5_keys, 1936 }; 1937 #endif 1938 1939 /* NOTE: A lot of things set to zero explicitly by call to 1940 * sk_alloc() so need not be done here. 1941 */ 1942 static int tcp_v6_init_sock(struct sock *sk) 1943 { 1944 struct inet_connection_sock *icsk = inet_csk(sk); 1945 1946 tcp_init_sock(sk); 1947 1948 icsk->icsk_af_ops = &ipv6_specific; 1949 1950 #ifdef CONFIG_TCP_MD5SIG 1951 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; 1952 #endif 1953 1954 return 0; 1955 } 1956 1957 #ifdef CONFIG_PROC_FS 1958 /* Proc filesystem TCPv6 sock list dumping. */ 1959 static void get_openreq6(struct seq_file *seq, 1960 const struct request_sock *req, int i) 1961 { 1962 long ttd = req->rsk_timer.expires - jiffies; 1963 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr; 1964 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr; 1965 1966 if (ttd < 0) 1967 ttd = 0; 1968 1969 seq_printf(seq, 1970 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1971 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", 1972 i, 1973 src->s6_addr32[0], src->s6_addr32[1], 1974 src->s6_addr32[2], src->s6_addr32[3], 1975 inet_rsk(req)->ir_num, 1976 dest->s6_addr32[0], dest->s6_addr32[1], 1977 dest->s6_addr32[2], dest->s6_addr32[3], 1978 ntohs(inet_rsk(req)->ir_rmt_port), 1979 TCP_SYN_RECV, 1980 0, 0, /* could print option size, but that is af dependent. */ 1981 1, /* timers active (only the expire timer) */ 1982 jiffies_to_clock_t(ttd), 1983 req->num_timeout, 1984 from_kuid_munged(seq_user_ns(seq), 1985 sock_i_uid(req->rsk_listener)), 1986 0, /* non standard timer */ 1987 0, /* open_requests have no inode */ 1988 0, req); 1989 } 1990 1991 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) 1992 { 1993 const struct in6_addr *dest, *src; 1994 __u16 destp, srcp; 1995 int timer_active; 1996 unsigned long timer_expires; 1997 const struct inet_sock *inet = inet_sk(sp); 1998 const struct tcp_sock *tp = tcp_sk(sp); 1999 const struct inet_connection_sock *icsk = inet_csk(sp); 2000 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; 2001 int rx_queue; 2002 int state; 2003 2004 dest = &sp->sk_v6_daddr; 2005 src = &sp->sk_v6_rcv_saddr; 2006 destp = ntohs(inet->inet_dport); 2007 srcp = ntohs(inet->inet_sport); 2008 2009 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 2010 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || 2011 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 2012 timer_active = 1; 2013 timer_expires = icsk->icsk_timeout; 2014 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 2015 timer_active = 4; 2016 timer_expires = icsk->icsk_timeout; 2017 } else if (timer_pending(&sp->sk_timer)) { 2018 timer_active = 2; 2019 timer_expires = sp->sk_timer.expires; 2020 } else { 2021 timer_active = 0; 2022 timer_expires = jiffies; 2023 } 2024 2025 state = inet_sk_state_load(sp); 2026 if (state == TCP_LISTEN) 2027 rx_queue = READ_ONCE(sp->sk_ack_backlog); 2028 else 2029 /* Because we don't lock the socket, 2030 * we might find a transient negative value. 2031 */ 2032 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) - 2033 READ_ONCE(tp->copied_seq), 0); 2034 2035 seq_printf(seq, 2036 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2037 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", 2038 i, 2039 src->s6_addr32[0], src->s6_addr32[1], 2040 src->s6_addr32[2], src->s6_addr32[3], srcp, 2041 dest->s6_addr32[0], dest->s6_addr32[1], 2042 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2043 state, 2044 READ_ONCE(tp->write_seq) - tp->snd_una, 2045 rx_queue, 2046 timer_active, 2047 jiffies_delta_to_clock_t(timer_expires - jiffies), 2048 icsk->icsk_retransmits, 2049 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)), 2050 icsk->icsk_probes_out, 2051 sock_i_ino(sp), 2052 refcount_read(&sp->sk_refcnt), sp, 2053 jiffies_to_clock_t(icsk->icsk_rto), 2054 jiffies_to_clock_t(icsk->icsk_ack.ato), 2055 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp), 2056 tcp_snd_cwnd(tp), 2057 state == TCP_LISTEN ? 2058 fastopenq->max_qlen : 2059 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) 2060 ); 2061 } 2062 2063 static void get_timewait6_sock(struct seq_file *seq, 2064 struct inet_timewait_sock *tw, int i) 2065 { 2066 long delta = tw->tw_timer.expires - jiffies; 2067 const struct in6_addr *dest, *src; 2068 __u16 destp, srcp; 2069 2070 dest = &tw->tw_v6_daddr; 2071 src = &tw->tw_v6_rcv_saddr; 2072 destp = ntohs(tw->tw_dport); 2073 srcp = ntohs(tw->tw_sport); 2074 2075 seq_printf(seq, 2076 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2077 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", 2078 i, 2079 src->s6_addr32[0], src->s6_addr32[1], 2080 src->s6_addr32[2], src->s6_addr32[3], srcp, 2081 dest->s6_addr32[0], dest->s6_addr32[1], 2082 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2083 tw->tw_substate, 0, 0, 2084 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2085 refcount_read(&tw->tw_refcnt), tw); 2086 } 2087 2088 static int tcp6_seq_show(struct seq_file *seq, void *v) 2089 { 2090 struct tcp_iter_state *st; 2091 struct sock *sk = v; 2092 2093 if (v == SEQ_START_TOKEN) { 2094 seq_puts(seq, 2095 " sl " 2096 "local_address " 2097 "remote_address " 2098 "st tx_queue rx_queue tr tm->when retrnsmt" 2099 " uid timeout inode\n"); 2100 goto out; 2101 } 2102 st = seq->private; 2103 2104 if (sk->sk_state == TCP_TIME_WAIT) 2105 get_timewait6_sock(seq, v, st->num); 2106 else if (sk->sk_state == TCP_NEW_SYN_RECV) 2107 get_openreq6(seq, v, st->num); 2108 else 2109 get_tcp6_sock(seq, v, st->num); 2110 out: 2111 return 0; 2112 } 2113 2114 static const struct seq_operations tcp6_seq_ops = { 2115 .show = tcp6_seq_show, 2116 .start = tcp_seq_start, 2117 .next = tcp_seq_next, 2118 .stop = tcp_seq_stop, 2119 }; 2120 2121 static struct tcp_seq_afinfo tcp6_seq_afinfo = { 2122 .family = AF_INET6, 2123 }; 2124 2125 int __net_init tcp6_proc_init(struct net *net) 2126 { 2127 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops, 2128 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo)) 2129 return -ENOMEM; 2130 return 0; 2131 } 2132 2133 void tcp6_proc_exit(struct net *net) 2134 { 2135 remove_proc_entry("tcp6", net->proc_net); 2136 } 2137 #endif 2138 2139 struct proto tcpv6_prot = { 2140 .name = "TCPv6", 2141 .owner = THIS_MODULE, 2142 .close = tcp_close, 2143 .pre_connect = tcp_v6_pre_connect, 2144 .connect = tcp_v6_connect, 2145 .disconnect = tcp_disconnect, 2146 .accept = inet_csk_accept, 2147 .ioctl = tcp_ioctl, 2148 .init = tcp_v6_init_sock, 2149 .destroy = tcp_v4_destroy_sock, 2150 .shutdown = tcp_shutdown, 2151 .setsockopt = tcp_setsockopt, 2152 .getsockopt = tcp_getsockopt, 2153 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt, 2154 .keepalive = tcp_set_keepalive, 2155 .recvmsg = tcp_recvmsg, 2156 .sendmsg = tcp_sendmsg, 2157 .splice_eof = tcp_splice_eof, 2158 .backlog_rcv = tcp_v6_do_rcv, 2159 .release_cb = tcp_release_cb, 2160 .hash = inet6_hash, 2161 .unhash = inet_unhash, 2162 .get_port = inet_csk_get_port, 2163 .put_port = inet_put_port, 2164 #ifdef CONFIG_BPF_SYSCALL 2165 .psock_update_sk_prot = tcp_bpf_update_proto, 2166 #endif 2167 .enter_memory_pressure = tcp_enter_memory_pressure, 2168 .leave_memory_pressure = tcp_leave_memory_pressure, 2169 .stream_memory_free = tcp_stream_memory_free, 2170 .sockets_allocated = &tcp_sockets_allocated, 2171 2172 .memory_allocated = &tcp_memory_allocated, 2173 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc, 2174 2175 .memory_pressure = &tcp_memory_pressure, 2176 .orphan_count = &tcp_orphan_count, 2177 .sysctl_mem = sysctl_tcp_mem, 2178 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), 2179 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), 2180 .max_header = MAX_TCP_HEADER, 2181 .obj_size = sizeof(struct tcp6_sock), 2182 .ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6), 2183 .slab_flags = SLAB_TYPESAFE_BY_RCU, 2184 .twsk_prot = &tcp6_timewait_sock_ops, 2185 .rsk_prot = &tcp6_request_sock_ops, 2186 .h.hashinfo = NULL, 2187 .no_autobind = true, 2188 .diag_destroy = tcp_abort, 2189 }; 2190 EXPORT_SYMBOL_GPL(tcpv6_prot); 2191 2192 static const struct inet6_protocol tcpv6_protocol = { 2193 .handler = tcp_v6_rcv, 2194 .err_handler = tcp_v6_err, 2195 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 2196 }; 2197 2198 static struct inet_protosw tcpv6_protosw = { 2199 .type = SOCK_STREAM, 2200 .protocol = IPPROTO_TCP, 2201 .prot = &tcpv6_prot, 2202 .ops = &inet6_stream_ops, 2203 .flags = INET_PROTOSW_PERMANENT | 2204 INET_PROTOSW_ICSK, 2205 }; 2206 2207 static int __net_init tcpv6_net_init(struct net *net) 2208 { 2209 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, 2210 SOCK_RAW, IPPROTO_TCP, net); 2211 } 2212 2213 static void __net_exit tcpv6_net_exit(struct net *net) 2214 { 2215 inet_ctl_sock_destroy(net->ipv6.tcp_sk); 2216 } 2217 2218 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list) 2219 { 2220 tcp_twsk_purge(net_exit_list, AF_INET6); 2221 } 2222 2223 static struct pernet_operations tcpv6_net_ops = { 2224 .init = tcpv6_net_init, 2225 .exit = tcpv6_net_exit, 2226 .exit_batch = tcpv6_net_exit_batch, 2227 }; 2228 2229 int __init tcpv6_init(void) 2230 { 2231 int ret; 2232 2233 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); 2234 if (ret) 2235 goto out; 2236 2237 /* register inet6 protocol */ 2238 ret = inet6_register_protosw(&tcpv6_protosw); 2239 if (ret) 2240 goto out_tcpv6_protocol; 2241 2242 ret = register_pernet_subsys(&tcpv6_net_ops); 2243 if (ret) 2244 goto out_tcpv6_protosw; 2245 2246 ret = mptcpv6_init(); 2247 if (ret) 2248 goto out_tcpv6_pernet_subsys; 2249 2250 out: 2251 return ret; 2252 2253 out_tcpv6_pernet_subsys: 2254 unregister_pernet_subsys(&tcpv6_net_ops); 2255 out_tcpv6_protosw: 2256 inet6_unregister_protosw(&tcpv6_protosw); 2257 out_tcpv6_protocol: 2258 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2259 goto out; 2260 } 2261 2262 void tcpv6_exit(void) 2263 { 2264 unregister_pernet_subsys(&tcpv6_net_ops); 2265 inet6_unregister_protosw(&tcpv6_protosw); 2266 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2267 } 2268