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.flowi6_oif = sk->sk_bound_dev_if; 276 fl6.flowi6_mark = sk->sk_mark; 277 fl6.fl6_dport = usin->sin6_port; 278 fl6.fl6_sport = inet->inet_sport; 279 fl6.flowi6_uid = sk->sk_uid; 280 281 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk)); 282 final_p = fl6_update_dst(&fl6, opt, &final); 283 284 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 285 286 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p); 287 if (IS_ERR(dst)) { 288 err = PTR_ERR(dst); 289 goto failure; 290 } 291 292 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; 293 294 if (!saddr) { 295 saddr = &fl6.saddr; 296 297 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6); 298 if (err) 299 goto failure; 300 } 301 302 /* set the source address */ 303 np->saddr = *saddr; 304 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 305 306 sk->sk_gso_type = SKB_GSO_TCPV6; 307 ip6_dst_store(sk, dst, NULL, NULL); 308 309 icsk->icsk_ext_hdr_len = 0; 310 if (opt) 311 icsk->icsk_ext_hdr_len = opt->opt_flen + 312 opt->opt_nflen; 313 314 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 315 316 inet->inet_dport = usin->sin6_port; 317 318 tcp_set_state(sk, TCP_SYN_SENT); 319 err = inet6_hash_connect(tcp_death_row, sk); 320 if (err) 321 goto late_failure; 322 323 sk_set_txhash(sk); 324 325 if (likely(!tp->repair)) { 326 if (!tp->write_seq) 327 WRITE_ONCE(tp->write_seq, 328 secure_tcpv6_seq(np->saddr.s6_addr32, 329 sk->sk_v6_daddr.s6_addr32, 330 inet->inet_sport, 331 inet->inet_dport)); 332 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32, 333 sk->sk_v6_daddr.s6_addr32); 334 } 335 336 if (tcp_fastopen_defer_connect(sk, &err)) 337 return err; 338 if (err) 339 goto late_failure; 340 341 err = tcp_connect(sk); 342 if (err) 343 goto late_failure; 344 345 return 0; 346 347 late_failure: 348 tcp_set_state(sk, TCP_CLOSE); 349 inet_bhash2_reset_saddr(sk); 350 failure: 351 inet->inet_dport = 0; 352 sk->sk_route_caps = 0; 353 return err; 354 } 355 356 static void tcp_v6_mtu_reduced(struct sock *sk) 357 { 358 struct dst_entry *dst; 359 u32 mtu; 360 361 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) 362 return; 363 364 mtu = READ_ONCE(tcp_sk(sk)->mtu_info); 365 366 /* Drop requests trying to increase our current mss. 367 * Check done in __ip6_rt_update_pmtu() is too late. 368 */ 369 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache) 370 return; 371 372 dst = inet6_csk_update_pmtu(sk, mtu); 373 if (!dst) 374 return; 375 376 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 377 tcp_sync_mss(sk, dst_mtu(dst)); 378 tcp_simple_retransmit(sk); 379 } 380 } 381 382 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 383 u8 type, u8 code, int offset, __be32 info) 384 { 385 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 386 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); 387 struct net *net = dev_net(skb->dev); 388 struct request_sock *fastopen; 389 struct ipv6_pinfo *np; 390 struct tcp_sock *tp; 391 __u32 seq, snd_una; 392 struct sock *sk; 393 bool fatal; 394 int err; 395 396 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 397 &hdr->daddr, th->dest, 398 &hdr->saddr, ntohs(th->source), 399 skb->dev->ifindex, inet6_sdif(skb)); 400 401 if (!sk) { 402 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 403 ICMP6_MIB_INERRORS); 404 return -ENOENT; 405 } 406 407 if (sk->sk_state == TCP_TIME_WAIT) { 408 inet_twsk_put(inet_twsk(sk)); 409 return 0; 410 } 411 seq = ntohl(th->seq); 412 fatal = icmpv6_err_convert(type, code, &err); 413 if (sk->sk_state == TCP_NEW_SYN_RECV) { 414 tcp_req_err(sk, seq, fatal); 415 return 0; 416 } 417 418 bh_lock_sock(sk); 419 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) 420 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 421 422 if (sk->sk_state == TCP_CLOSE) 423 goto out; 424 425 if (static_branch_unlikely(&ip6_min_hopcount)) { 426 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ 427 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) { 428 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 429 goto out; 430 } 431 } 432 433 tp = tcp_sk(sk); 434 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ 435 fastopen = rcu_dereference(tp->fastopen_rsk); 436 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; 437 if (sk->sk_state != TCP_LISTEN && 438 !between(seq, snd_una, tp->snd_nxt)) { 439 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 440 goto out; 441 } 442 443 np = tcp_inet6_sk(sk); 444 445 if (type == NDISC_REDIRECT) { 446 if (!sock_owned_by_user(sk)) { 447 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie); 448 449 if (dst) 450 dst->ops->redirect(dst, sk, skb); 451 } 452 goto out; 453 } 454 455 if (type == ICMPV6_PKT_TOOBIG) { 456 u32 mtu = ntohl(info); 457 458 /* We are not interested in TCP_LISTEN and open_requests 459 * (SYN-ACKs send out by Linux are always <576bytes so 460 * they should go through unfragmented). 461 */ 462 if (sk->sk_state == TCP_LISTEN) 463 goto out; 464 465 if (!ip6_sk_accept_pmtu(sk)) 466 goto out; 467 468 if (mtu < IPV6_MIN_MTU) 469 goto out; 470 471 WRITE_ONCE(tp->mtu_info, mtu); 472 473 if (!sock_owned_by_user(sk)) 474 tcp_v6_mtu_reduced(sk); 475 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, 476 &sk->sk_tsq_flags)) 477 sock_hold(sk); 478 goto out; 479 } 480 481 482 /* Might be for an request_sock */ 483 switch (sk->sk_state) { 484 case TCP_SYN_SENT: 485 case TCP_SYN_RECV: 486 /* Only in fast or simultaneous open. If a fast open socket is 487 * already accepted it is treated as a connected one below. 488 */ 489 if (fastopen && !fastopen->sk) 490 break; 491 492 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th); 493 494 if (!sock_owned_by_user(sk)) { 495 sk->sk_err = err; 496 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ 497 498 tcp_done(sk); 499 } else 500 sk->sk_err_soft = err; 501 goto out; 502 case TCP_LISTEN: 503 break; 504 default: 505 /* check if this ICMP message allows revert of backoff. 506 * (see RFC 6069) 507 */ 508 if (!fastopen && type == ICMPV6_DEST_UNREACH && 509 code == ICMPV6_NOROUTE) 510 tcp_ld_RTO_revert(sk, seq); 511 } 512 513 if (!sock_owned_by_user(sk) && np->recverr) { 514 sk->sk_err = err; 515 sk_error_report(sk); 516 } else 517 sk->sk_err_soft = err; 518 519 out: 520 bh_unlock_sock(sk); 521 sock_put(sk); 522 return 0; 523 } 524 525 526 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, 527 struct flowi *fl, 528 struct request_sock *req, 529 struct tcp_fastopen_cookie *foc, 530 enum tcp_synack_type synack_type, 531 struct sk_buff *syn_skb) 532 { 533 struct inet_request_sock *ireq = inet_rsk(req); 534 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 535 struct ipv6_txoptions *opt; 536 struct flowi6 *fl6 = &fl->u.ip6; 537 struct sk_buff *skb; 538 int err = -ENOMEM; 539 u8 tclass; 540 541 /* First, grab a route. */ 542 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req, 543 IPPROTO_TCP)) == NULL) 544 goto done; 545 546 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb); 547 548 if (skb) { 549 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, 550 &ireq->ir_v6_rmt_addr); 551 552 fl6->daddr = ireq->ir_v6_rmt_addr; 553 if (np->repflow && ireq->pktopts) 554 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); 555 556 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ? 557 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) | 558 (np->tclass & INET_ECN_MASK) : 559 np->tclass; 560 561 if (!INET_ECN_is_capable(tclass) && 562 tcp_bpf_ca_needs_ecn((struct sock *)req)) 563 tclass |= INET_ECN_ECT_0; 564 565 rcu_read_lock(); 566 opt = ireq->ipv6_opt; 567 if (!opt) 568 opt = rcu_dereference(np->opt); 569 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt, 570 tclass, sk->sk_priority); 571 rcu_read_unlock(); 572 err = net_xmit_eval(err); 573 } 574 575 done: 576 return err; 577 } 578 579 580 static void tcp_v6_reqsk_destructor(struct request_sock *req) 581 { 582 kfree(inet_rsk(req)->ipv6_opt); 583 consume_skb(inet_rsk(req)->pktopts); 584 } 585 586 #ifdef CONFIG_TCP_MD5SIG 587 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, 588 const struct in6_addr *addr, 589 int l3index) 590 { 591 return tcp_md5_do_lookup(sk, l3index, 592 (union tcp_md5_addr *)addr, AF_INET6); 593 } 594 595 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk, 596 const struct sock *addr_sk) 597 { 598 int l3index; 599 600 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 601 addr_sk->sk_bound_dev_if); 602 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr, 603 l3index); 604 } 605 606 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname, 607 sockptr_t optval, int optlen) 608 { 609 struct tcp_md5sig cmd; 610 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; 611 int l3index = 0; 612 u8 prefixlen; 613 u8 flags; 614 615 if (optlen < sizeof(cmd)) 616 return -EINVAL; 617 618 if (copy_from_sockptr(&cmd, optval, sizeof(cmd))) 619 return -EFAULT; 620 621 if (sin6->sin6_family != AF_INET6) 622 return -EINVAL; 623 624 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX; 625 626 if (optname == TCP_MD5SIG_EXT && 627 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { 628 prefixlen = cmd.tcpm_prefixlen; 629 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) && 630 prefixlen > 32)) 631 return -EINVAL; 632 } else { 633 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128; 634 } 635 636 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex && 637 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) { 638 struct net_device *dev; 639 640 rcu_read_lock(); 641 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex); 642 if (dev && netif_is_l3_master(dev)) 643 l3index = dev->ifindex; 644 rcu_read_unlock(); 645 646 /* ok to reference set/not set outside of rcu; 647 * right now device MUST be an L3 master 648 */ 649 if (!dev || !l3index) 650 return -EINVAL; 651 } 652 653 if (!cmd.tcpm_keylen) { 654 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 655 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 656 AF_INET, prefixlen, 657 l3index, flags); 658 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 659 AF_INET6, prefixlen, l3index, flags); 660 } 661 662 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 663 return -EINVAL; 664 665 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 666 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 667 AF_INET, prefixlen, l3index, flags, 668 cmd.tcpm_key, cmd.tcpm_keylen); 669 670 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 671 AF_INET6, prefixlen, l3index, flags, 672 cmd.tcpm_key, cmd.tcpm_keylen); 673 } 674 675 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp, 676 const struct in6_addr *daddr, 677 const struct in6_addr *saddr, 678 const struct tcphdr *th, int nbytes) 679 { 680 struct tcp6_pseudohdr *bp; 681 struct scatterlist sg; 682 struct tcphdr *_th; 683 684 bp = hp->scratch; 685 /* 1. TCP pseudo-header (RFC2460) */ 686 bp->saddr = *saddr; 687 bp->daddr = *daddr; 688 bp->protocol = cpu_to_be32(IPPROTO_TCP); 689 bp->len = cpu_to_be32(nbytes); 690 691 _th = (struct tcphdr *)(bp + 1); 692 memcpy(_th, th, sizeof(*th)); 693 _th->check = 0; 694 695 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); 696 ahash_request_set_crypt(hp->md5_req, &sg, NULL, 697 sizeof(*bp) + sizeof(*th)); 698 return crypto_ahash_update(hp->md5_req); 699 } 700 701 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 702 const struct in6_addr *daddr, struct in6_addr *saddr, 703 const struct tcphdr *th) 704 { 705 struct tcp_md5sig_pool *hp; 706 struct ahash_request *req; 707 708 hp = tcp_get_md5sig_pool(); 709 if (!hp) 710 goto clear_hash_noput; 711 req = hp->md5_req; 712 713 if (crypto_ahash_init(req)) 714 goto clear_hash; 715 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) 716 goto clear_hash; 717 if (tcp_md5_hash_key(hp, key)) 718 goto clear_hash; 719 ahash_request_set_crypt(req, NULL, md5_hash, 0); 720 if (crypto_ahash_final(req)) 721 goto clear_hash; 722 723 tcp_put_md5sig_pool(); 724 return 0; 725 726 clear_hash: 727 tcp_put_md5sig_pool(); 728 clear_hash_noput: 729 memset(md5_hash, 0, 16); 730 return 1; 731 } 732 733 static int tcp_v6_md5_hash_skb(char *md5_hash, 734 const struct tcp_md5sig_key *key, 735 const struct sock *sk, 736 const struct sk_buff *skb) 737 { 738 const struct in6_addr *saddr, *daddr; 739 struct tcp_md5sig_pool *hp; 740 struct ahash_request *req; 741 const struct tcphdr *th = tcp_hdr(skb); 742 743 if (sk) { /* valid for establish/request sockets */ 744 saddr = &sk->sk_v6_rcv_saddr; 745 daddr = &sk->sk_v6_daddr; 746 } else { 747 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 748 saddr = &ip6h->saddr; 749 daddr = &ip6h->daddr; 750 } 751 752 hp = tcp_get_md5sig_pool(); 753 if (!hp) 754 goto clear_hash_noput; 755 req = hp->md5_req; 756 757 if (crypto_ahash_init(req)) 758 goto clear_hash; 759 760 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len)) 761 goto clear_hash; 762 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 763 goto clear_hash; 764 if (tcp_md5_hash_key(hp, key)) 765 goto clear_hash; 766 ahash_request_set_crypt(req, NULL, md5_hash, 0); 767 if (crypto_ahash_final(req)) 768 goto clear_hash; 769 770 tcp_put_md5sig_pool(); 771 return 0; 772 773 clear_hash: 774 tcp_put_md5sig_pool(); 775 clear_hash_noput: 776 memset(md5_hash, 0, 16); 777 return 1; 778 } 779 780 #endif 781 782 static void tcp_v6_init_req(struct request_sock *req, 783 const struct sock *sk_listener, 784 struct sk_buff *skb) 785 { 786 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 787 struct inet_request_sock *ireq = inet_rsk(req); 788 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener); 789 790 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; 791 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; 792 793 /* So that link locals have meaning */ 794 if ((!sk_listener->sk_bound_dev_if || l3_slave) && 795 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) 796 ireq->ir_iif = tcp_v6_iif(skb); 797 798 if (!TCP_SKB_CB(skb)->tcp_tw_isn && 799 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) || 800 np->rxopt.bits.rxinfo || 801 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || 802 np->rxopt.bits.rxohlim || np->repflow)) { 803 refcount_inc(&skb->users); 804 ireq->pktopts = skb; 805 } 806 } 807 808 static struct dst_entry *tcp_v6_route_req(const struct sock *sk, 809 struct sk_buff *skb, 810 struct flowi *fl, 811 struct request_sock *req) 812 { 813 tcp_v6_init_req(req, sk, skb); 814 815 if (security_inet_conn_request(sk, skb, req)) 816 return NULL; 817 818 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP); 819 } 820 821 struct request_sock_ops tcp6_request_sock_ops __read_mostly = { 822 .family = AF_INET6, 823 .obj_size = sizeof(struct tcp6_request_sock), 824 .rtx_syn_ack = tcp_rtx_synack, 825 .send_ack = tcp_v6_reqsk_send_ack, 826 .destructor = tcp_v6_reqsk_destructor, 827 .send_reset = tcp_v6_send_reset, 828 .syn_ack_timeout = tcp_syn_ack_timeout, 829 }; 830 831 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { 832 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 833 sizeof(struct ipv6hdr), 834 #ifdef CONFIG_TCP_MD5SIG 835 .req_md5_lookup = tcp_v6_md5_lookup, 836 .calc_md5_hash = tcp_v6_md5_hash_skb, 837 #endif 838 #ifdef CONFIG_SYN_COOKIES 839 .cookie_init_seq = cookie_v6_init_sequence, 840 #endif 841 .route_req = tcp_v6_route_req, 842 .init_seq = tcp_v6_init_seq, 843 .init_ts_off = tcp_v6_init_ts_off, 844 .send_synack = tcp_v6_send_synack, 845 }; 846 847 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, 848 u32 ack, u32 win, u32 tsval, u32 tsecr, 849 int oif, struct tcp_md5sig_key *key, int rst, 850 u8 tclass, __be32 label, u32 priority, u32 txhash) 851 { 852 const struct tcphdr *th = tcp_hdr(skb); 853 struct tcphdr *t1; 854 struct sk_buff *buff; 855 struct flowi6 fl6; 856 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 857 struct sock *ctl_sk = net->ipv6.tcp_sk; 858 unsigned int tot_len = sizeof(struct tcphdr); 859 __be32 mrst = 0, *topt; 860 struct dst_entry *dst; 861 __u32 mark = 0; 862 863 if (tsecr) 864 tot_len += TCPOLEN_TSTAMP_ALIGNED; 865 #ifdef CONFIG_TCP_MD5SIG 866 if (key) 867 tot_len += TCPOLEN_MD5SIG_ALIGNED; 868 #endif 869 870 #ifdef CONFIG_MPTCP 871 if (rst && !key) { 872 mrst = mptcp_reset_option(skb); 873 874 if (mrst) 875 tot_len += sizeof(__be32); 876 } 877 #endif 878 879 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC); 880 if (!buff) 881 return; 882 883 skb_reserve(buff, MAX_TCP_HEADER); 884 885 t1 = skb_push(buff, tot_len); 886 skb_reset_transport_header(buff); 887 888 /* Swap the send and the receive. */ 889 memset(t1, 0, sizeof(*t1)); 890 t1->dest = th->source; 891 t1->source = th->dest; 892 t1->doff = tot_len / 4; 893 t1->seq = htonl(seq); 894 t1->ack_seq = htonl(ack); 895 t1->ack = !rst || !th->ack; 896 t1->rst = rst; 897 t1->window = htons(win); 898 899 topt = (__be32 *)(t1 + 1); 900 901 if (tsecr) { 902 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 903 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 904 *topt++ = htonl(tsval); 905 *topt++ = htonl(tsecr); 906 } 907 908 if (mrst) 909 *topt++ = mrst; 910 911 #ifdef CONFIG_TCP_MD5SIG 912 if (key) { 913 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 914 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); 915 tcp_v6_md5_hash_hdr((__u8 *)topt, key, 916 &ipv6_hdr(skb)->saddr, 917 &ipv6_hdr(skb)->daddr, t1); 918 } 919 #endif 920 921 memset(&fl6, 0, sizeof(fl6)); 922 fl6.daddr = ipv6_hdr(skb)->saddr; 923 fl6.saddr = ipv6_hdr(skb)->daddr; 924 fl6.flowlabel = label; 925 926 buff->ip_summed = CHECKSUM_PARTIAL; 927 928 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr); 929 930 fl6.flowi6_proto = IPPROTO_TCP; 931 if (rt6_need_strict(&fl6.daddr) && !oif) 932 fl6.flowi6_oif = tcp_v6_iif(skb); 933 else { 934 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 935 oif = skb->skb_iif; 936 937 fl6.flowi6_oif = oif; 938 } 939 940 if (sk) { 941 if (sk->sk_state == TCP_TIME_WAIT) 942 mark = inet_twsk(sk)->tw_mark; 943 else 944 mark = sk->sk_mark; 945 skb_set_delivery_time(buff, tcp_transmit_time(sk), true); 946 } 947 if (txhash) { 948 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */ 949 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4); 950 } 951 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark; 952 fl6.fl6_dport = t1->dest; 953 fl6.fl6_sport = t1->source; 954 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); 955 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6)); 956 957 /* Pass a socket to ip6_dst_lookup either it is for RST 958 * Underlying function will use this to retrieve the network 959 * namespace 960 */ 961 if (sk && sk->sk_state != TCP_TIME_WAIT) 962 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/ 963 else 964 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL); 965 if (!IS_ERR(dst)) { 966 skb_dst_set(buff, dst); 967 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, 968 tclass & ~INET_ECN_MASK, priority); 969 TCP_INC_STATS(net, TCP_MIB_OUTSEGS); 970 if (rst) 971 TCP_INC_STATS(net, TCP_MIB_OUTRSTS); 972 return; 973 } 974 975 kfree_skb(buff); 976 } 977 978 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) 979 { 980 const struct tcphdr *th = tcp_hdr(skb); 981 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 982 u32 seq = 0, ack_seq = 0; 983 struct tcp_md5sig_key *key = NULL; 984 #ifdef CONFIG_TCP_MD5SIG 985 const __u8 *hash_location = NULL; 986 unsigned char newhash[16]; 987 int genhash; 988 struct sock *sk1 = NULL; 989 #endif 990 __be32 label = 0; 991 u32 priority = 0; 992 struct net *net; 993 u32 txhash = 0; 994 int oif = 0; 995 996 if (th->rst) 997 return; 998 999 /* If sk not NULL, it means we did a successful lookup and incoming 1000 * route had to be correct. prequeue might have dropped our dst. 1001 */ 1002 if (!sk && !ipv6_unicast_destination(skb)) 1003 return; 1004 1005 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 1006 #ifdef CONFIG_TCP_MD5SIG 1007 rcu_read_lock(); 1008 hash_location = tcp_parse_md5sig_option(th); 1009 if (sk && sk_fullsock(sk)) { 1010 int l3index; 1011 1012 /* sdif set, means packet ingressed via a device 1013 * in an L3 domain and inet_iif is set to it. 1014 */ 1015 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1016 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index); 1017 } else if (hash_location) { 1018 int dif = tcp_v6_iif_l3_slave(skb); 1019 int sdif = tcp_v6_sdif(skb); 1020 int l3index; 1021 1022 /* 1023 * active side is lost. Try to find listening socket through 1024 * source port, and then find md5 key through listening socket. 1025 * we are not loose security here: 1026 * Incoming packet is checked with md5 hash with finding key, 1027 * no RST generated if md5 hash doesn't match. 1028 */ 1029 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1030 NULL, 0, &ipv6h->saddr, th->source, 1031 &ipv6h->daddr, ntohs(th->source), 1032 dif, sdif); 1033 if (!sk1) 1034 goto out; 1035 1036 /* sdif set, means packet ingressed via a device 1037 * in an L3 domain and dif is set to it. 1038 */ 1039 l3index = tcp_v6_sdif(skb) ? dif : 0; 1040 1041 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index); 1042 if (!key) 1043 goto out; 1044 1045 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb); 1046 if (genhash || memcmp(hash_location, newhash, 16) != 0) 1047 goto out; 1048 } 1049 #endif 1050 1051 if (th->ack) 1052 seq = ntohl(th->ack_seq); 1053 else 1054 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - 1055 (th->doff << 2); 1056 1057 if (sk) { 1058 oif = sk->sk_bound_dev_if; 1059 if (sk_fullsock(sk)) { 1060 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1061 1062 trace_tcp_send_reset(sk, skb); 1063 if (np->repflow) 1064 label = ip6_flowlabel(ipv6h); 1065 priority = sk->sk_priority; 1066 txhash = sk->sk_hash; 1067 } 1068 if (sk->sk_state == TCP_TIME_WAIT) { 1069 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel); 1070 priority = inet_twsk(sk)->tw_priority; 1071 txhash = inet_twsk(sk)->tw_txhash; 1072 } 1073 } else { 1074 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET) 1075 label = ip6_flowlabel(ipv6h); 1076 } 1077 1078 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 1079 ipv6_get_dsfield(ipv6h), label, priority, txhash); 1080 1081 #ifdef CONFIG_TCP_MD5SIG 1082 out: 1083 rcu_read_unlock(); 1084 #endif 1085 } 1086 1087 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, 1088 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, 1089 struct tcp_md5sig_key *key, u8 tclass, 1090 __be32 label, u32 priority, u32 txhash) 1091 { 1092 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0, 1093 tclass, label, priority, txhash); 1094 } 1095 1096 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) 1097 { 1098 struct inet_timewait_sock *tw = inet_twsk(sk); 1099 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 1100 1101 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 1102 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 1103 tcp_time_stamp_raw() + tcptw->tw_ts_offset, 1104 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw), 1105 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority, 1106 tw->tw_txhash); 1107 1108 inet_twsk_put(tw); 1109 } 1110 1111 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 1112 struct request_sock *req) 1113 { 1114 int l3index; 1115 1116 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1117 1118 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV 1119 * sk->sk_state == TCP_SYN_RECV -> for Fast Open. 1120 */ 1121 /* RFC 7323 2.3 1122 * The window field (SEG.WND) of every outgoing segment, with the 1123 * exception of <SYN> segments, MUST be right-shifted by 1124 * Rcv.Wind.Shift bits: 1125 */ 1126 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ? 1127 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, 1128 tcp_rsk(req)->rcv_nxt, 1129 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, 1130 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off, 1131 req->ts_recent, sk->sk_bound_dev_if, 1132 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index), 1133 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority, 1134 tcp_rsk(req)->txhash); 1135 } 1136 1137 1138 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) 1139 { 1140 #ifdef CONFIG_SYN_COOKIES 1141 const struct tcphdr *th = tcp_hdr(skb); 1142 1143 if (!th->syn) 1144 sk = cookie_v6_check(sk, skb); 1145 #endif 1146 return sk; 1147 } 1148 1149 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 1150 struct tcphdr *th, u32 *cookie) 1151 { 1152 u16 mss = 0; 1153 #ifdef CONFIG_SYN_COOKIES 1154 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops, 1155 &tcp_request_sock_ipv6_ops, sk, th); 1156 if (mss) { 1157 *cookie = __cookie_v6_init_sequence(iph, th, &mss); 1158 tcp_synq_overflow(sk); 1159 } 1160 #endif 1161 return mss; 1162 } 1163 1164 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 1165 { 1166 if (skb->protocol == htons(ETH_P_IP)) 1167 return tcp_v4_conn_request(sk, skb); 1168 1169 if (!ipv6_unicast_destination(skb)) 1170 goto drop; 1171 1172 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) { 1173 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); 1174 return 0; 1175 } 1176 1177 return tcp_conn_request(&tcp6_request_sock_ops, 1178 &tcp_request_sock_ipv6_ops, sk, skb); 1179 1180 drop: 1181 tcp_listendrop(sk); 1182 return 0; /* don't send reset */ 1183 } 1184 1185 static void tcp_v6_restore_cb(struct sk_buff *skb) 1186 { 1187 /* We need to move header back to the beginning if xfrm6_policy_check() 1188 * and tcp_v6_fill_cb() are going to be called again. 1189 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there. 1190 */ 1191 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, 1192 sizeof(struct inet6_skb_parm)); 1193 } 1194 1195 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 1196 struct request_sock *req, 1197 struct dst_entry *dst, 1198 struct request_sock *req_unhash, 1199 bool *own_req) 1200 { 1201 struct inet_request_sock *ireq; 1202 struct ipv6_pinfo *newnp; 1203 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1204 struct ipv6_txoptions *opt; 1205 struct inet_sock *newinet; 1206 bool found_dup_sk = false; 1207 struct tcp_sock *newtp; 1208 struct sock *newsk; 1209 #ifdef CONFIG_TCP_MD5SIG 1210 struct tcp_md5sig_key *key; 1211 int l3index; 1212 #endif 1213 struct flowi6 fl6; 1214 1215 if (skb->protocol == htons(ETH_P_IP)) { 1216 /* 1217 * v6 mapped 1218 */ 1219 1220 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, 1221 req_unhash, own_req); 1222 1223 if (!newsk) 1224 return NULL; 1225 1226 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1227 1228 newnp = tcp_inet6_sk(newsk); 1229 newtp = tcp_sk(newsk); 1230 1231 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1232 1233 newnp->saddr = newsk->sk_v6_rcv_saddr; 1234 1235 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; 1236 if (sk_is_mptcp(newsk)) 1237 mptcpv6_handle_mapped(newsk, true); 1238 newsk->sk_backlog_rcv = tcp_v4_do_rcv; 1239 #ifdef CONFIG_TCP_MD5SIG 1240 newtp->af_specific = &tcp_sock_ipv6_mapped_specific; 1241 #endif 1242 1243 newnp->ipv6_mc_list = NULL; 1244 newnp->ipv6_ac_list = NULL; 1245 newnp->ipv6_fl_list = NULL; 1246 newnp->pktoptions = NULL; 1247 newnp->opt = NULL; 1248 newnp->mcast_oif = inet_iif(skb); 1249 newnp->mcast_hops = ip_hdr(skb)->ttl; 1250 newnp->rcv_flowinfo = 0; 1251 if (np->repflow) 1252 newnp->flow_label = 0; 1253 1254 /* 1255 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 1256 * here, tcp_create_openreq_child now does this for us, see the comment in 1257 * that function for the gory details. -acme 1258 */ 1259 1260 /* It is tricky place. Until this moment IPv4 tcp 1261 worked with IPv6 icsk.icsk_af_ops. 1262 Sync it now. 1263 */ 1264 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 1265 1266 return newsk; 1267 } 1268 1269 ireq = inet_rsk(req); 1270 1271 if (sk_acceptq_is_full(sk)) 1272 goto out_overflow; 1273 1274 if (!dst) { 1275 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP); 1276 if (!dst) 1277 goto out; 1278 } 1279 1280 newsk = tcp_create_openreq_child(sk, req, skb); 1281 if (!newsk) 1282 goto out_nonewsk; 1283 1284 /* 1285 * No need to charge this sock to the relevant IPv6 refcnt debug socks 1286 * count here, tcp_create_openreq_child now does this for us, see the 1287 * comment in that function for the gory details. -acme 1288 */ 1289 1290 newsk->sk_gso_type = SKB_GSO_TCPV6; 1291 ip6_dst_store(newsk, dst, NULL, NULL); 1292 inet6_sk_rx_dst_set(newsk, skb); 1293 1294 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1295 1296 newtp = tcp_sk(newsk); 1297 newinet = inet_sk(newsk); 1298 newnp = tcp_inet6_sk(newsk); 1299 1300 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1301 1302 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr; 1303 newnp->saddr = ireq->ir_v6_loc_addr; 1304 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr; 1305 newsk->sk_bound_dev_if = ireq->ir_iif; 1306 1307 /* Now IPv6 options... 1308 1309 First: no IPv4 options. 1310 */ 1311 newinet->inet_opt = NULL; 1312 newnp->ipv6_mc_list = NULL; 1313 newnp->ipv6_ac_list = NULL; 1314 newnp->ipv6_fl_list = NULL; 1315 1316 /* Clone RX bits */ 1317 newnp->rxopt.all = np->rxopt.all; 1318 1319 newnp->pktoptions = NULL; 1320 newnp->opt = NULL; 1321 newnp->mcast_oif = tcp_v6_iif(skb); 1322 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1323 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); 1324 if (np->repflow) 1325 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); 1326 1327 /* Set ToS of the new socket based upon the value of incoming SYN. 1328 * ECT bits are set later in tcp_init_transfer(). 1329 */ 1330 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)) 1331 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK; 1332 1333 /* Clone native IPv6 options from listening socket (if any) 1334 1335 Yes, keeping reference count would be much more clever, 1336 but we make one more one thing there: reattach optmem 1337 to newsk. 1338 */ 1339 opt = ireq->ipv6_opt; 1340 if (!opt) 1341 opt = rcu_dereference(np->opt); 1342 if (opt) { 1343 opt = ipv6_dup_options(newsk, opt); 1344 RCU_INIT_POINTER(newnp->opt, opt); 1345 } 1346 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1347 if (opt) 1348 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen + 1349 opt->opt_flen; 1350 1351 tcp_ca_openreq_child(newsk, dst); 1352 1353 tcp_sync_mss(newsk, dst_mtu(dst)); 1354 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst)); 1355 1356 tcp_initialize_rcv_mss(newsk); 1357 1358 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 1359 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 1360 1361 #ifdef CONFIG_TCP_MD5SIG 1362 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif); 1363 1364 /* Copy over the MD5 key from the original socket */ 1365 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index); 1366 if (key) { 1367 const union tcp_md5_addr *addr; 1368 1369 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr; 1370 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) { 1371 inet_csk_prepare_forced_close(newsk); 1372 tcp_done(newsk); 1373 goto out; 1374 } 1375 } 1376 #endif 1377 1378 if (__inet_inherit_port(sk, newsk) < 0) { 1379 inet_csk_prepare_forced_close(newsk); 1380 tcp_done(newsk); 1381 goto out; 1382 } 1383 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), 1384 &found_dup_sk); 1385 if (*own_req) { 1386 tcp_move_syn(newtp, req); 1387 1388 /* Clone pktoptions received with SYN, if we own the req */ 1389 if (ireq->pktopts) { 1390 newnp->pktoptions = skb_clone(ireq->pktopts, 1391 sk_gfp_mask(sk, GFP_ATOMIC)); 1392 consume_skb(ireq->pktopts); 1393 ireq->pktopts = NULL; 1394 if (newnp->pktoptions) { 1395 tcp_v6_restore_cb(newnp->pktoptions); 1396 skb_set_owner_r(newnp->pktoptions, newsk); 1397 } 1398 } 1399 } else { 1400 if (!req_unhash && found_dup_sk) { 1401 /* This code path should only be executed in the 1402 * syncookie case only 1403 */ 1404 bh_unlock_sock(newsk); 1405 sock_put(newsk); 1406 newsk = NULL; 1407 } 1408 } 1409 1410 return newsk; 1411 1412 out_overflow: 1413 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1414 out_nonewsk: 1415 dst_release(dst); 1416 out: 1417 tcp_listendrop(sk); 1418 return NULL; 1419 } 1420 1421 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 1422 u32)); 1423 /* The socket must have it's spinlock held when we get 1424 * here, unless it is a TCP_LISTEN socket. 1425 * 1426 * We have a potential double-lock case here, so even when 1427 * doing backlog processing we use the BH locking scheme. 1428 * This is because we cannot sleep with the original spinlock 1429 * held. 1430 */ 1431 INDIRECT_CALLABLE_SCOPE 1432 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 1433 { 1434 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1435 struct sk_buff *opt_skb = NULL; 1436 enum skb_drop_reason reason; 1437 struct tcp_sock *tp; 1438 1439 /* Imagine: socket is IPv6. IPv4 packet arrives, 1440 goes to IPv4 receive handler and backlogged. 1441 From backlog it always goes here. Kerboom... 1442 Fortunately, tcp_rcv_established and rcv_established 1443 handle them correctly, but it is not case with 1444 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK 1445 */ 1446 1447 if (skb->protocol == htons(ETH_P_IP)) 1448 return tcp_v4_do_rcv(sk, skb); 1449 1450 /* 1451 * socket locking is here for SMP purposes as backlog rcv 1452 * is currently called with bh processing disabled. 1453 */ 1454 1455 /* Do Stevens' IPV6_PKTOPTIONS. 1456 1457 Yes, guys, it is the only place in our code, where we 1458 may make it not affecting IPv4. 1459 The rest of code is protocol independent, 1460 and I do not like idea to uglify IPv4. 1461 1462 Actually, all the idea behind IPV6_PKTOPTIONS 1463 looks not very well thought. For now we latch 1464 options, received in the last packet, enqueued 1465 by tcp. Feel free to propose better solution. 1466 --ANK (980728) 1467 */ 1468 if (np->rxopt.all) 1469 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 1470 1471 reason = SKB_DROP_REASON_NOT_SPECIFIED; 1472 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1473 struct dst_entry *dst; 1474 1475 dst = rcu_dereference_protected(sk->sk_rx_dst, 1476 lockdep_sock_is_held(sk)); 1477 1478 sock_rps_save_rxhash(sk, skb); 1479 sk_mark_napi_id(sk, skb); 1480 if (dst) { 1481 if (sk->sk_rx_dst_ifindex != skb->skb_iif || 1482 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check, 1483 dst, sk->sk_rx_dst_cookie) == NULL) { 1484 RCU_INIT_POINTER(sk->sk_rx_dst, NULL); 1485 dst_release(dst); 1486 } 1487 } 1488 1489 tcp_rcv_established(sk, skb); 1490 if (opt_skb) 1491 goto ipv6_pktoptions; 1492 return 0; 1493 } 1494 1495 if (tcp_checksum_complete(skb)) 1496 goto csum_err; 1497 1498 if (sk->sk_state == TCP_LISTEN) { 1499 struct sock *nsk = tcp_v6_cookie_check(sk, skb); 1500 1501 if (!nsk) 1502 goto discard; 1503 1504 if (nsk != sk) { 1505 if (tcp_child_process(sk, nsk, skb)) 1506 goto reset; 1507 if (opt_skb) 1508 __kfree_skb(opt_skb); 1509 return 0; 1510 } 1511 } else 1512 sock_rps_save_rxhash(sk, skb); 1513 1514 if (tcp_rcv_state_process(sk, skb)) 1515 goto reset; 1516 if (opt_skb) 1517 goto ipv6_pktoptions; 1518 return 0; 1519 1520 reset: 1521 tcp_v6_send_reset(sk, skb); 1522 discard: 1523 if (opt_skb) 1524 __kfree_skb(opt_skb); 1525 kfree_skb_reason(skb, reason); 1526 return 0; 1527 csum_err: 1528 reason = SKB_DROP_REASON_TCP_CSUM; 1529 trace_tcp_bad_csum(skb); 1530 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); 1531 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); 1532 goto discard; 1533 1534 1535 ipv6_pktoptions: 1536 /* Do you ask, what is it? 1537 1538 1. skb was enqueued by tcp. 1539 2. skb is added to tail of read queue, rather than out of order. 1540 3. socket is not in passive state. 1541 4. Finally, it really contains options, which user wants to receive. 1542 */ 1543 tp = tcp_sk(sk); 1544 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && 1545 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { 1546 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) 1547 np->mcast_oif = tcp_v6_iif(opt_skb); 1548 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) 1549 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit; 1550 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) 1551 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb)); 1552 if (np->repflow) 1553 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb)); 1554 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) { 1555 skb_set_owner_r(opt_skb, sk); 1556 tcp_v6_restore_cb(opt_skb); 1557 opt_skb = xchg(&np->pktoptions, opt_skb); 1558 } else { 1559 __kfree_skb(opt_skb); 1560 opt_skb = xchg(&np->pktoptions, NULL); 1561 } 1562 } 1563 1564 consume_skb(opt_skb); 1565 return 0; 1566 } 1567 1568 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, 1569 const struct tcphdr *th) 1570 { 1571 /* This is tricky: we move IP6CB at its correct location into 1572 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because 1573 * _decode_session6() uses IP6CB(). 1574 * barrier() makes sure compiler won't play aliasing games. 1575 */ 1576 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), 1577 sizeof(struct inet6_skb_parm)); 1578 barrier(); 1579 1580 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1581 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1582 skb->len - th->doff*4); 1583 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1584 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1585 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1586 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr); 1587 TCP_SKB_CB(skb)->sacked = 0; 1588 TCP_SKB_CB(skb)->has_rxtstamp = 1589 skb->tstamp || skb_hwtstamps(skb)->hwtstamp; 1590 } 1591 1592 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) 1593 { 1594 enum skb_drop_reason drop_reason; 1595 int sdif = inet6_sdif(skb); 1596 int dif = inet6_iif(skb); 1597 const struct tcphdr *th; 1598 const struct ipv6hdr *hdr; 1599 bool refcounted; 1600 struct sock *sk; 1601 int ret; 1602 struct net *net = dev_net(skb->dev); 1603 1604 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 1605 if (skb->pkt_type != PACKET_HOST) 1606 goto discard_it; 1607 1608 /* 1609 * Count it even if it's bad. 1610 */ 1611 __TCP_INC_STATS(net, TCP_MIB_INSEGS); 1612 1613 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1614 goto discard_it; 1615 1616 th = (const struct tcphdr *)skb->data; 1617 1618 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) { 1619 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1620 goto bad_packet; 1621 } 1622 if (!pskb_may_pull(skb, th->doff*4)) 1623 goto discard_it; 1624 1625 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) 1626 goto csum_error; 1627 1628 th = (const struct tcphdr *)skb->data; 1629 hdr = ipv6_hdr(skb); 1630 1631 lookup: 1632 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th), 1633 th->source, th->dest, inet6_iif(skb), sdif, 1634 &refcounted); 1635 if (!sk) 1636 goto no_tcp_socket; 1637 1638 process: 1639 if (sk->sk_state == TCP_TIME_WAIT) 1640 goto do_time_wait; 1641 1642 if (sk->sk_state == TCP_NEW_SYN_RECV) { 1643 struct request_sock *req = inet_reqsk(sk); 1644 bool req_stolen = false; 1645 struct sock *nsk; 1646 1647 sk = req->rsk_listener; 1648 drop_reason = tcp_inbound_md5_hash(sk, skb, 1649 &hdr->saddr, &hdr->daddr, 1650 AF_INET6, dif, sdif); 1651 if (drop_reason) { 1652 sk_drops_add(sk, skb); 1653 reqsk_put(req); 1654 goto discard_it; 1655 } 1656 if (tcp_checksum_complete(skb)) { 1657 reqsk_put(req); 1658 goto csum_error; 1659 } 1660 if (unlikely(sk->sk_state != TCP_LISTEN)) { 1661 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb); 1662 if (!nsk) { 1663 inet_csk_reqsk_queue_drop_and_put(sk, req); 1664 goto lookup; 1665 } 1666 sk = nsk; 1667 /* reuseport_migrate_sock() has already held one sk_refcnt 1668 * before returning. 1669 */ 1670 } else { 1671 sock_hold(sk); 1672 } 1673 refcounted = true; 1674 nsk = NULL; 1675 if (!tcp_filter(sk, skb)) { 1676 th = (const struct tcphdr *)skb->data; 1677 hdr = ipv6_hdr(skb); 1678 tcp_v6_fill_cb(skb, hdr, th); 1679 nsk = tcp_check_req(sk, skb, req, false, &req_stolen); 1680 } else { 1681 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1682 } 1683 if (!nsk) { 1684 reqsk_put(req); 1685 if (req_stolen) { 1686 /* Another cpu got exclusive access to req 1687 * and created a full blown socket. 1688 * Try to feed this packet to this socket 1689 * instead of discarding it. 1690 */ 1691 tcp_v6_restore_cb(skb); 1692 sock_put(sk); 1693 goto lookup; 1694 } 1695 goto discard_and_relse; 1696 } 1697 if (nsk == sk) { 1698 reqsk_put(req); 1699 tcp_v6_restore_cb(skb); 1700 } else if (tcp_child_process(sk, nsk, skb)) { 1701 tcp_v6_send_reset(nsk, skb); 1702 goto discard_and_relse; 1703 } else { 1704 sock_put(sk); 1705 return 0; 1706 } 1707 } 1708 1709 if (static_branch_unlikely(&ip6_min_hopcount)) { 1710 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ 1711 if (hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) { 1712 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 1713 goto discard_and_relse; 1714 } 1715 } 1716 1717 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 1718 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1719 goto discard_and_relse; 1720 } 1721 1722 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr, 1723 AF_INET6, dif, sdif); 1724 if (drop_reason) 1725 goto discard_and_relse; 1726 1727 if (tcp_filter(sk, skb)) { 1728 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 1729 goto discard_and_relse; 1730 } 1731 th = (const struct tcphdr *)skb->data; 1732 hdr = ipv6_hdr(skb); 1733 tcp_v6_fill_cb(skb, hdr, th); 1734 1735 skb->dev = NULL; 1736 1737 if (sk->sk_state == TCP_LISTEN) { 1738 ret = tcp_v6_do_rcv(sk, skb); 1739 goto put_and_return; 1740 } 1741 1742 sk_incoming_cpu_update(sk); 1743 1744 bh_lock_sock_nested(sk); 1745 tcp_segs_in(tcp_sk(sk), skb); 1746 ret = 0; 1747 if (!sock_owned_by_user(sk)) { 1748 ret = tcp_v6_do_rcv(sk, skb); 1749 } else { 1750 if (tcp_add_backlog(sk, skb, &drop_reason)) 1751 goto discard_and_relse; 1752 } 1753 bh_unlock_sock(sk); 1754 put_and_return: 1755 if (refcounted) 1756 sock_put(sk); 1757 return ret ? -1 : 0; 1758 1759 no_tcp_socket: 1760 drop_reason = SKB_DROP_REASON_NO_SOCKET; 1761 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1762 goto discard_it; 1763 1764 tcp_v6_fill_cb(skb, hdr, th); 1765 1766 if (tcp_checksum_complete(skb)) { 1767 csum_error: 1768 drop_reason = SKB_DROP_REASON_TCP_CSUM; 1769 trace_tcp_bad_csum(skb); 1770 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); 1771 bad_packet: 1772 __TCP_INC_STATS(net, TCP_MIB_INERRS); 1773 } else { 1774 tcp_v6_send_reset(NULL, skb); 1775 } 1776 1777 discard_it: 1778 SKB_DR_OR(drop_reason, NOT_SPECIFIED); 1779 kfree_skb_reason(skb, drop_reason); 1780 return 0; 1781 1782 discard_and_relse: 1783 sk_drops_add(sk, skb); 1784 if (refcounted) 1785 sock_put(sk); 1786 goto discard_it; 1787 1788 do_time_wait: 1789 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1790 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 1791 inet_twsk_put(inet_twsk(sk)); 1792 goto discard_it; 1793 } 1794 1795 tcp_v6_fill_cb(skb, hdr, th); 1796 1797 if (tcp_checksum_complete(skb)) { 1798 inet_twsk_put(inet_twsk(sk)); 1799 goto csum_error; 1800 } 1801 1802 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1803 case TCP_TW_SYN: 1804 { 1805 struct sock *sk2; 1806 1807 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo, 1808 skb, __tcp_hdrlen(th), 1809 &ipv6_hdr(skb)->saddr, th->source, 1810 &ipv6_hdr(skb)->daddr, 1811 ntohs(th->dest), 1812 tcp_v6_iif_l3_slave(skb), 1813 sdif); 1814 if (sk2) { 1815 struct inet_timewait_sock *tw = inet_twsk(sk); 1816 inet_twsk_deschedule_put(tw); 1817 sk = sk2; 1818 tcp_v6_restore_cb(skb); 1819 refcounted = false; 1820 goto process; 1821 } 1822 } 1823 /* to ACK */ 1824 fallthrough; 1825 case TCP_TW_ACK: 1826 tcp_v6_timewait_ack(sk, skb); 1827 break; 1828 case TCP_TW_RST: 1829 tcp_v6_send_reset(sk, skb); 1830 inet_twsk_deschedule_put(inet_twsk(sk)); 1831 goto discard_it; 1832 case TCP_TW_SUCCESS: 1833 ; 1834 } 1835 goto discard_it; 1836 } 1837 1838 void tcp_v6_early_demux(struct sk_buff *skb) 1839 { 1840 struct net *net = dev_net(skb->dev); 1841 const struct ipv6hdr *hdr; 1842 const struct tcphdr *th; 1843 struct sock *sk; 1844 1845 if (skb->pkt_type != PACKET_HOST) 1846 return; 1847 1848 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1849 return; 1850 1851 hdr = ipv6_hdr(skb); 1852 th = tcp_hdr(skb); 1853 1854 if (th->doff < sizeof(struct tcphdr) / 4) 1855 return; 1856 1857 /* Note : We use inet6_iif() here, not tcp_v6_iif() */ 1858 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, 1859 &hdr->saddr, th->source, 1860 &hdr->daddr, ntohs(th->dest), 1861 inet6_iif(skb), inet6_sdif(skb)); 1862 if (sk) { 1863 skb->sk = sk; 1864 skb->destructor = sock_edemux; 1865 if (sk_fullsock(sk)) { 1866 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst); 1867 1868 if (dst) 1869 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1870 if (dst && 1871 sk->sk_rx_dst_ifindex == skb->skb_iif) 1872 skb_dst_set_noref(skb, dst); 1873 } 1874 } 1875 } 1876 1877 static struct timewait_sock_ops tcp6_timewait_sock_ops = { 1878 .twsk_obj_size = sizeof(struct tcp6_timewait_sock), 1879 .twsk_unique = tcp_twsk_unique, 1880 .twsk_destructor = tcp_twsk_destructor, 1881 }; 1882 1883 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb) 1884 { 1885 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr); 1886 } 1887 1888 const struct inet_connection_sock_af_ops ipv6_specific = { 1889 .queue_xmit = inet6_csk_xmit, 1890 .send_check = tcp_v6_send_check, 1891 .rebuild_header = inet6_sk_rebuild_header, 1892 .sk_rx_dst_set = inet6_sk_rx_dst_set, 1893 .conn_request = tcp_v6_conn_request, 1894 .syn_recv_sock = tcp_v6_syn_recv_sock, 1895 .net_header_len = sizeof(struct ipv6hdr), 1896 .net_frag_header_len = sizeof(struct frag_hdr), 1897 .setsockopt = ipv6_setsockopt, 1898 .getsockopt = ipv6_getsockopt, 1899 .addr2sockaddr = inet6_csk_addr2sockaddr, 1900 .sockaddr_len = sizeof(struct sockaddr_in6), 1901 .mtu_reduced = tcp_v6_mtu_reduced, 1902 }; 1903 1904 #ifdef CONFIG_TCP_MD5SIG 1905 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { 1906 .md5_lookup = tcp_v6_md5_lookup, 1907 .calc_md5_hash = tcp_v6_md5_hash_skb, 1908 .md5_parse = tcp_v6_parse_md5_keys, 1909 }; 1910 #endif 1911 1912 /* 1913 * TCP over IPv4 via INET6 API 1914 */ 1915 static const struct inet_connection_sock_af_ops ipv6_mapped = { 1916 .queue_xmit = ip_queue_xmit, 1917 .send_check = tcp_v4_send_check, 1918 .rebuild_header = inet_sk_rebuild_header, 1919 .sk_rx_dst_set = inet_sk_rx_dst_set, 1920 .conn_request = tcp_v6_conn_request, 1921 .syn_recv_sock = tcp_v6_syn_recv_sock, 1922 .net_header_len = sizeof(struct iphdr), 1923 .setsockopt = ipv6_setsockopt, 1924 .getsockopt = ipv6_getsockopt, 1925 .addr2sockaddr = inet6_csk_addr2sockaddr, 1926 .sockaddr_len = sizeof(struct sockaddr_in6), 1927 .mtu_reduced = tcp_v4_mtu_reduced, 1928 }; 1929 1930 #ifdef CONFIG_TCP_MD5SIG 1931 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { 1932 .md5_lookup = tcp_v4_md5_lookup, 1933 .calc_md5_hash = tcp_v4_md5_hash_skb, 1934 .md5_parse = tcp_v6_parse_md5_keys, 1935 }; 1936 #endif 1937 1938 /* NOTE: A lot of things set to zero explicitly by call to 1939 * sk_alloc() so need not be done here. 1940 */ 1941 static int tcp_v6_init_sock(struct sock *sk) 1942 { 1943 struct inet_connection_sock *icsk = inet_csk(sk); 1944 1945 tcp_init_sock(sk); 1946 1947 icsk->icsk_af_ops = &ipv6_specific; 1948 1949 #ifdef CONFIG_TCP_MD5SIG 1950 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; 1951 #endif 1952 1953 return 0; 1954 } 1955 1956 #ifdef CONFIG_PROC_FS 1957 /* Proc filesystem TCPv6 sock list dumping. */ 1958 static void get_openreq6(struct seq_file *seq, 1959 const struct request_sock *req, int i) 1960 { 1961 long ttd = req->rsk_timer.expires - jiffies; 1962 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr; 1963 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr; 1964 1965 if (ttd < 0) 1966 ttd = 0; 1967 1968 seq_printf(seq, 1969 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1970 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", 1971 i, 1972 src->s6_addr32[0], src->s6_addr32[1], 1973 src->s6_addr32[2], src->s6_addr32[3], 1974 inet_rsk(req)->ir_num, 1975 dest->s6_addr32[0], dest->s6_addr32[1], 1976 dest->s6_addr32[2], dest->s6_addr32[3], 1977 ntohs(inet_rsk(req)->ir_rmt_port), 1978 TCP_SYN_RECV, 1979 0, 0, /* could print option size, but that is af dependent. */ 1980 1, /* timers active (only the expire timer) */ 1981 jiffies_to_clock_t(ttd), 1982 req->num_timeout, 1983 from_kuid_munged(seq_user_ns(seq), 1984 sock_i_uid(req->rsk_listener)), 1985 0, /* non standard timer */ 1986 0, /* open_requests have no inode */ 1987 0, req); 1988 } 1989 1990 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) 1991 { 1992 const struct in6_addr *dest, *src; 1993 __u16 destp, srcp; 1994 int timer_active; 1995 unsigned long timer_expires; 1996 const struct inet_sock *inet = inet_sk(sp); 1997 const struct tcp_sock *tp = tcp_sk(sp); 1998 const struct inet_connection_sock *icsk = inet_csk(sp); 1999 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; 2000 int rx_queue; 2001 int state; 2002 2003 dest = &sp->sk_v6_daddr; 2004 src = &sp->sk_v6_rcv_saddr; 2005 destp = ntohs(inet->inet_dport); 2006 srcp = ntohs(inet->inet_sport); 2007 2008 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 2009 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || 2010 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 2011 timer_active = 1; 2012 timer_expires = icsk->icsk_timeout; 2013 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 2014 timer_active = 4; 2015 timer_expires = icsk->icsk_timeout; 2016 } else if (timer_pending(&sp->sk_timer)) { 2017 timer_active = 2; 2018 timer_expires = sp->sk_timer.expires; 2019 } else { 2020 timer_active = 0; 2021 timer_expires = jiffies; 2022 } 2023 2024 state = inet_sk_state_load(sp); 2025 if (state == TCP_LISTEN) 2026 rx_queue = READ_ONCE(sp->sk_ack_backlog); 2027 else 2028 /* Because we don't lock the socket, 2029 * we might find a transient negative value. 2030 */ 2031 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) - 2032 READ_ONCE(tp->copied_seq), 0); 2033 2034 seq_printf(seq, 2035 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2036 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", 2037 i, 2038 src->s6_addr32[0], src->s6_addr32[1], 2039 src->s6_addr32[2], src->s6_addr32[3], srcp, 2040 dest->s6_addr32[0], dest->s6_addr32[1], 2041 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2042 state, 2043 READ_ONCE(tp->write_seq) - tp->snd_una, 2044 rx_queue, 2045 timer_active, 2046 jiffies_delta_to_clock_t(timer_expires - jiffies), 2047 icsk->icsk_retransmits, 2048 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)), 2049 icsk->icsk_probes_out, 2050 sock_i_ino(sp), 2051 refcount_read(&sp->sk_refcnt), sp, 2052 jiffies_to_clock_t(icsk->icsk_rto), 2053 jiffies_to_clock_t(icsk->icsk_ack.ato), 2054 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp), 2055 tcp_snd_cwnd(tp), 2056 state == TCP_LISTEN ? 2057 fastopenq->max_qlen : 2058 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) 2059 ); 2060 } 2061 2062 static void get_timewait6_sock(struct seq_file *seq, 2063 struct inet_timewait_sock *tw, int i) 2064 { 2065 long delta = tw->tw_timer.expires - jiffies; 2066 const struct in6_addr *dest, *src; 2067 __u16 destp, srcp; 2068 2069 dest = &tw->tw_v6_daddr; 2070 src = &tw->tw_v6_rcv_saddr; 2071 destp = ntohs(tw->tw_dport); 2072 srcp = ntohs(tw->tw_sport); 2073 2074 seq_printf(seq, 2075 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2076 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", 2077 i, 2078 src->s6_addr32[0], src->s6_addr32[1], 2079 src->s6_addr32[2], src->s6_addr32[3], srcp, 2080 dest->s6_addr32[0], dest->s6_addr32[1], 2081 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2082 tw->tw_substate, 0, 0, 2083 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2084 refcount_read(&tw->tw_refcnt), tw); 2085 } 2086 2087 static int tcp6_seq_show(struct seq_file *seq, void *v) 2088 { 2089 struct tcp_iter_state *st; 2090 struct sock *sk = v; 2091 2092 if (v == SEQ_START_TOKEN) { 2093 seq_puts(seq, 2094 " sl " 2095 "local_address " 2096 "remote_address " 2097 "st tx_queue rx_queue tr tm->when retrnsmt" 2098 " uid timeout inode\n"); 2099 goto out; 2100 } 2101 st = seq->private; 2102 2103 if (sk->sk_state == TCP_TIME_WAIT) 2104 get_timewait6_sock(seq, v, st->num); 2105 else if (sk->sk_state == TCP_NEW_SYN_RECV) 2106 get_openreq6(seq, v, st->num); 2107 else 2108 get_tcp6_sock(seq, v, st->num); 2109 out: 2110 return 0; 2111 } 2112 2113 static const struct seq_operations tcp6_seq_ops = { 2114 .show = tcp6_seq_show, 2115 .start = tcp_seq_start, 2116 .next = tcp_seq_next, 2117 .stop = tcp_seq_stop, 2118 }; 2119 2120 static struct tcp_seq_afinfo tcp6_seq_afinfo = { 2121 .family = AF_INET6, 2122 }; 2123 2124 int __net_init tcp6_proc_init(struct net *net) 2125 { 2126 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops, 2127 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo)) 2128 return -ENOMEM; 2129 return 0; 2130 } 2131 2132 void tcp6_proc_exit(struct net *net) 2133 { 2134 remove_proc_entry("tcp6", net->proc_net); 2135 } 2136 #endif 2137 2138 struct proto tcpv6_prot = { 2139 .name = "TCPv6", 2140 .owner = THIS_MODULE, 2141 .close = tcp_close, 2142 .pre_connect = tcp_v6_pre_connect, 2143 .connect = tcp_v6_connect, 2144 .disconnect = tcp_disconnect, 2145 .accept = inet_csk_accept, 2146 .ioctl = tcp_ioctl, 2147 .init = tcp_v6_init_sock, 2148 .destroy = tcp_v4_destroy_sock, 2149 .shutdown = tcp_shutdown, 2150 .setsockopt = tcp_setsockopt, 2151 .getsockopt = tcp_getsockopt, 2152 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt, 2153 .keepalive = tcp_set_keepalive, 2154 .recvmsg = tcp_recvmsg, 2155 .sendmsg = tcp_sendmsg, 2156 .sendpage = tcp_sendpage, 2157 .backlog_rcv = tcp_v6_do_rcv, 2158 .release_cb = tcp_release_cb, 2159 .hash = inet6_hash, 2160 .unhash = inet_unhash, 2161 .get_port = inet_csk_get_port, 2162 .put_port = inet_put_port, 2163 #ifdef CONFIG_BPF_SYSCALL 2164 .psock_update_sk_prot = tcp_bpf_update_proto, 2165 #endif 2166 .enter_memory_pressure = tcp_enter_memory_pressure, 2167 .leave_memory_pressure = tcp_leave_memory_pressure, 2168 .stream_memory_free = tcp_stream_memory_free, 2169 .sockets_allocated = &tcp_sockets_allocated, 2170 2171 .memory_allocated = &tcp_memory_allocated, 2172 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc, 2173 2174 .memory_pressure = &tcp_memory_pressure, 2175 .orphan_count = &tcp_orphan_count, 2176 .sysctl_mem = sysctl_tcp_mem, 2177 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), 2178 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), 2179 .max_header = MAX_TCP_HEADER, 2180 .obj_size = sizeof(struct tcp6_sock), 2181 .slab_flags = SLAB_TYPESAFE_BY_RCU, 2182 .twsk_prot = &tcp6_timewait_sock_ops, 2183 .rsk_prot = &tcp6_request_sock_ops, 2184 .h.hashinfo = NULL, 2185 .no_autobind = true, 2186 .diag_destroy = tcp_abort, 2187 }; 2188 EXPORT_SYMBOL_GPL(tcpv6_prot); 2189 2190 static const struct inet6_protocol tcpv6_protocol = { 2191 .handler = tcp_v6_rcv, 2192 .err_handler = tcp_v6_err, 2193 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 2194 }; 2195 2196 static struct inet_protosw tcpv6_protosw = { 2197 .type = SOCK_STREAM, 2198 .protocol = IPPROTO_TCP, 2199 .prot = &tcpv6_prot, 2200 .ops = &inet6_stream_ops, 2201 .flags = INET_PROTOSW_PERMANENT | 2202 INET_PROTOSW_ICSK, 2203 }; 2204 2205 static int __net_init tcpv6_net_init(struct net *net) 2206 { 2207 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, 2208 SOCK_RAW, IPPROTO_TCP, net); 2209 } 2210 2211 static void __net_exit tcpv6_net_exit(struct net *net) 2212 { 2213 inet_ctl_sock_destroy(net->ipv6.tcp_sk); 2214 } 2215 2216 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list) 2217 { 2218 tcp_twsk_purge(net_exit_list, AF_INET6); 2219 } 2220 2221 static struct pernet_operations tcpv6_net_ops = { 2222 .init = tcpv6_net_init, 2223 .exit = tcpv6_net_exit, 2224 .exit_batch = tcpv6_net_exit_batch, 2225 }; 2226 2227 int __init tcpv6_init(void) 2228 { 2229 int ret; 2230 2231 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); 2232 if (ret) 2233 goto out; 2234 2235 /* register inet6 protocol */ 2236 ret = inet6_register_protosw(&tcpv6_protosw); 2237 if (ret) 2238 goto out_tcpv6_protocol; 2239 2240 ret = register_pernet_subsys(&tcpv6_net_ops); 2241 if (ret) 2242 goto out_tcpv6_protosw; 2243 2244 ret = mptcpv6_init(); 2245 if (ret) 2246 goto out_tcpv6_pernet_subsys; 2247 2248 out: 2249 return ret; 2250 2251 out_tcpv6_pernet_subsys: 2252 unregister_pernet_subsys(&tcpv6_net_ops); 2253 out_tcpv6_protosw: 2254 inet6_unregister_protosw(&tcpv6_protosw); 2255 out_tcpv6_protocol: 2256 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2257 goto out; 2258 } 2259 2260 void tcpv6_exit(void) 2261 { 2262 unregister_pernet_subsys(&tcpv6_net_ops); 2263 inet6_unregister_protosw(&tcpv6_protosw); 2264 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2265 } 2266