1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * TCP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on: 10 * linux/net/ipv4/tcp.c 11 * linux/net/ipv4/tcp_input.c 12 * linux/net/ipv4/tcp_output.c 13 * 14 * Fixes: 15 * Hideaki YOSHIFUJI : sin6_scope_id support 16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 18 * a single port at the same time. 19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file. 20 */ 21 22 #include <linux/bottom_half.h> 23 #include <linux/module.h> 24 #include <linux/errno.h> 25 #include <linux/types.h> 26 #include <linux/socket.h> 27 #include <linux/sockios.h> 28 #include <linux/net.h> 29 #include <linux/jiffies.h> 30 #include <linux/in.h> 31 #include <linux/in6.h> 32 #include <linux/netdevice.h> 33 #include <linux/init.h> 34 #include <linux/jhash.h> 35 #include <linux/ipsec.h> 36 #include <linux/times.h> 37 #include <linux/slab.h> 38 #include <linux/uaccess.h> 39 #include <linux/ipv6.h> 40 #include <linux/icmpv6.h> 41 #include <linux/random.h> 42 #include <linux/indirect_call_wrapper.h> 43 44 #include <net/tcp.h> 45 #include <net/ndisc.h> 46 #include <net/inet6_hashtables.h> 47 #include <net/inet6_connection_sock.h> 48 #include <net/ipv6.h> 49 #include <net/transp_v6.h> 50 #include <net/addrconf.h> 51 #include <net/ip6_route.h> 52 #include <net/ip6_checksum.h> 53 #include <net/inet_ecn.h> 54 #include <net/protocol.h> 55 #include <net/xfrm.h> 56 #include <net/snmp.h> 57 #include <net/dsfield.h> 58 #include <net/timewait_sock.h> 59 #include <net/inet_common.h> 60 #include <net/secure_seq.h> 61 #include <net/busy_poll.h> 62 63 #include <linux/proc_fs.h> 64 #include <linux/seq_file.h> 65 66 #include <crypto/hash.h> 67 #include <linux/scatterlist.h> 68 69 #include <trace/events/tcp.h> 70 71 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb); 72 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 73 struct request_sock *req); 74 75 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb); 76 77 static const struct inet_connection_sock_af_ops ipv6_mapped; 78 const struct inet_connection_sock_af_ops ipv6_specific; 79 #ifdef CONFIG_TCP_MD5SIG 80 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific; 81 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific; 82 #else 83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, 84 const struct in6_addr *addr, 85 int l3index) 86 { 87 return NULL; 88 } 89 #endif 90 91 /* Helper returning the inet6 address from a given tcp socket. 92 * It can be used in TCP stack instead of inet6_sk(sk). 93 * This avoids a dereference and allow compiler optimizations. 94 * It is a specialized version of inet6_sk_generic(). 95 */ 96 static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk) 97 { 98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo); 99 100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset); 101 } 102 103 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) 104 { 105 struct dst_entry *dst = skb_dst(skb); 106 107 if (dst && dst_hold_safe(dst)) { 108 const struct rt6_info *rt = (const struct rt6_info *)dst; 109 110 sk->sk_rx_dst = dst; 111 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; 112 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 113 } 114 } 115 116 static u32 tcp_v6_init_seq(const struct sk_buff *skb) 117 { 118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32, 119 ipv6_hdr(skb)->saddr.s6_addr32, 120 tcp_hdr(skb)->dest, 121 tcp_hdr(skb)->source); 122 } 123 124 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb) 125 { 126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32, 127 ipv6_hdr(skb)->saddr.s6_addr32); 128 } 129 130 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 131 int addr_len) 132 { 133 /* This check is replicated from tcp_v6_connect() and intended to 134 * prevent BPF program called below from accessing bytes that are out 135 * of the bound specified by user in addr_len. 136 */ 137 if (addr_len < SIN6_LEN_RFC2133) 138 return -EINVAL; 139 140 sock_owned_by_me(sk); 141 142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr); 143 } 144 145 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr, 146 int addr_len) 147 { 148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr; 149 struct inet_sock *inet = inet_sk(sk); 150 struct inet_connection_sock *icsk = inet_csk(sk); 151 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 152 struct tcp_sock *tp = tcp_sk(sk); 153 struct in6_addr *saddr = NULL, *final_p, final; 154 struct ipv6_txoptions *opt; 155 struct flowi6 fl6; 156 struct dst_entry *dst; 157 int addr_type; 158 int err; 159 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; 160 161 if (addr_len < SIN6_LEN_RFC2133) 162 return -EINVAL; 163 164 if (usin->sin6_family != AF_INET6) 165 return -EAFNOSUPPORT; 166 167 memset(&fl6, 0, sizeof(fl6)); 168 169 if (np->sndflow) { 170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; 171 IP6_ECN_flow_init(fl6.flowlabel); 172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 173 struct ip6_flowlabel *flowlabel; 174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 175 if (IS_ERR(flowlabel)) 176 return -EINVAL; 177 fl6_sock_release(flowlabel); 178 } 179 } 180 181 /* 182 * connect() to INADDR_ANY means loopback (BSD'ism). 183 */ 184 185 if (ipv6_addr_any(&usin->sin6_addr)) { 186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 188 &usin->sin6_addr); 189 else 190 usin->sin6_addr = in6addr_loopback; 191 } 192 193 addr_type = ipv6_addr_type(&usin->sin6_addr); 194 195 if (addr_type & IPV6_ADDR_MULTICAST) 196 return -ENETUNREACH; 197 198 if (addr_type&IPV6_ADDR_LINKLOCAL) { 199 if (addr_len >= sizeof(struct sockaddr_in6) && 200 usin->sin6_scope_id) { 201 /* If interface is set while binding, indices 202 * must coincide. 203 */ 204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id)) 205 return -EINVAL; 206 207 sk->sk_bound_dev_if = usin->sin6_scope_id; 208 } 209 210 /* Connect to link-local address requires an interface */ 211 if (!sk->sk_bound_dev_if) 212 return -EINVAL; 213 } 214 215 if (tp->rx_opt.ts_recent_stamp && 216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) { 217 tp->rx_opt.ts_recent = 0; 218 tp->rx_opt.ts_recent_stamp = 0; 219 WRITE_ONCE(tp->write_seq, 0); 220 } 221 222 sk->sk_v6_daddr = usin->sin6_addr; 223 np->flow_label = fl6.flowlabel; 224 225 /* 226 * TCP over IPv4 227 */ 228 229 if (addr_type & IPV6_ADDR_MAPPED) { 230 u32 exthdrlen = icsk->icsk_ext_hdr_len; 231 struct sockaddr_in sin; 232 233 if (__ipv6_only_sock(sk)) 234 return -ENETUNREACH; 235 236 sin.sin_family = AF_INET; 237 sin.sin_port = usin->sin6_port; 238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 239 240 icsk->icsk_af_ops = &ipv6_mapped; 241 if (sk_is_mptcp(sk)) 242 mptcpv6_handle_mapped(sk, true); 243 sk->sk_backlog_rcv = tcp_v4_do_rcv; 244 #ifdef CONFIG_TCP_MD5SIG 245 tp->af_specific = &tcp_sock_ipv6_mapped_specific; 246 #endif 247 248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 249 250 if (err) { 251 icsk->icsk_ext_hdr_len = exthdrlen; 252 icsk->icsk_af_ops = &ipv6_specific; 253 if (sk_is_mptcp(sk)) 254 mptcpv6_handle_mapped(sk, false); 255 sk->sk_backlog_rcv = tcp_v6_do_rcv; 256 #ifdef CONFIG_TCP_MD5SIG 257 tp->af_specific = &tcp_sock_ipv6_specific; 258 #endif 259 goto failure; 260 } 261 np->saddr = sk->sk_v6_rcv_saddr; 262 263 return err; 264 } 265 266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 267 saddr = &sk->sk_v6_rcv_saddr; 268 269 fl6.flowi6_proto = IPPROTO_TCP; 270 fl6.daddr = sk->sk_v6_daddr; 271 fl6.saddr = saddr ? *saddr : np->saddr; 272 fl6.flowi6_oif = sk->sk_bound_dev_if; 273 fl6.flowi6_mark = sk->sk_mark; 274 fl6.fl6_dport = usin->sin6_port; 275 fl6.fl6_sport = inet->inet_sport; 276 fl6.flowi6_uid = sk->sk_uid; 277 278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk)); 279 final_p = fl6_update_dst(&fl6, opt, &final); 280 281 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 282 283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); 284 if (IS_ERR(dst)) { 285 err = PTR_ERR(dst); 286 goto failure; 287 } 288 289 if (!saddr) { 290 saddr = &fl6.saddr; 291 sk->sk_v6_rcv_saddr = *saddr; 292 } 293 294 /* set the source address */ 295 np->saddr = *saddr; 296 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 297 298 sk->sk_gso_type = SKB_GSO_TCPV6; 299 ip6_dst_store(sk, dst, NULL, NULL); 300 301 icsk->icsk_ext_hdr_len = 0; 302 if (opt) 303 icsk->icsk_ext_hdr_len = opt->opt_flen + 304 opt->opt_nflen; 305 306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 307 308 inet->inet_dport = usin->sin6_port; 309 310 tcp_set_state(sk, TCP_SYN_SENT); 311 err = inet6_hash_connect(tcp_death_row, sk); 312 if (err) 313 goto late_failure; 314 315 sk_set_txhash(sk); 316 317 if (likely(!tp->repair)) { 318 if (!tp->write_seq) 319 WRITE_ONCE(tp->write_seq, 320 secure_tcpv6_seq(np->saddr.s6_addr32, 321 sk->sk_v6_daddr.s6_addr32, 322 inet->inet_sport, 323 inet->inet_dport)); 324 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk), 325 np->saddr.s6_addr32, 326 sk->sk_v6_daddr.s6_addr32); 327 } 328 329 if (tcp_fastopen_defer_connect(sk, &err)) 330 return err; 331 if (err) 332 goto late_failure; 333 334 err = tcp_connect(sk); 335 if (err) 336 goto late_failure; 337 338 return 0; 339 340 late_failure: 341 tcp_set_state(sk, TCP_CLOSE); 342 failure: 343 inet->inet_dport = 0; 344 sk->sk_route_caps = 0; 345 return err; 346 } 347 348 static void tcp_v6_mtu_reduced(struct sock *sk) 349 { 350 struct dst_entry *dst; 351 352 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) 353 return; 354 355 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info); 356 if (!dst) 357 return; 358 359 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 360 tcp_sync_mss(sk, dst_mtu(dst)); 361 tcp_simple_retransmit(sk); 362 } 363 } 364 365 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 366 u8 type, u8 code, int offset, __be32 info) 367 { 368 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 369 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); 370 struct net *net = dev_net(skb->dev); 371 struct request_sock *fastopen; 372 struct ipv6_pinfo *np; 373 struct tcp_sock *tp; 374 __u32 seq, snd_una; 375 struct sock *sk; 376 bool fatal; 377 int err; 378 379 sk = __inet6_lookup_established(net, &tcp_hashinfo, 380 &hdr->daddr, th->dest, 381 &hdr->saddr, ntohs(th->source), 382 skb->dev->ifindex, inet6_sdif(skb)); 383 384 if (!sk) { 385 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 386 ICMP6_MIB_INERRORS); 387 return -ENOENT; 388 } 389 390 if (sk->sk_state == TCP_TIME_WAIT) { 391 inet_twsk_put(inet_twsk(sk)); 392 return 0; 393 } 394 seq = ntohl(th->seq); 395 fatal = icmpv6_err_convert(type, code, &err); 396 if (sk->sk_state == TCP_NEW_SYN_RECV) { 397 tcp_req_err(sk, seq, fatal); 398 return 0; 399 } 400 401 bh_lock_sock(sk); 402 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) 403 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 404 405 if (sk->sk_state == TCP_CLOSE) 406 goto out; 407 408 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) { 409 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 410 goto out; 411 } 412 413 tp = tcp_sk(sk); 414 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ 415 fastopen = rcu_dereference(tp->fastopen_rsk); 416 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; 417 if (sk->sk_state != TCP_LISTEN && 418 !between(seq, snd_una, tp->snd_nxt)) { 419 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 420 goto out; 421 } 422 423 np = tcp_inet6_sk(sk); 424 425 if (type == NDISC_REDIRECT) { 426 if (!sock_owned_by_user(sk)) { 427 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie); 428 429 if (dst) 430 dst->ops->redirect(dst, sk, skb); 431 } 432 goto out; 433 } 434 435 if (type == ICMPV6_PKT_TOOBIG) { 436 /* We are not interested in TCP_LISTEN and open_requests 437 * (SYN-ACKs send out by Linux are always <576bytes so 438 * they should go through unfragmented). 439 */ 440 if (sk->sk_state == TCP_LISTEN) 441 goto out; 442 443 if (!ip6_sk_accept_pmtu(sk)) 444 goto out; 445 446 tp->mtu_info = ntohl(info); 447 if (!sock_owned_by_user(sk)) 448 tcp_v6_mtu_reduced(sk); 449 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, 450 &sk->sk_tsq_flags)) 451 sock_hold(sk); 452 goto out; 453 } 454 455 456 /* Might be for an request_sock */ 457 switch (sk->sk_state) { 458 case TCP_SYN_SENT: 459 case TCP_SYN_RECV: 460 /* Only in fast or simultaneous open. If a fast open socket is 461 * is already accepted it is treated as a connected one below. 462 */ 463 if (fastopen && !fastopen->sk) 464 break; 465 466 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th); 467 468 if (!sock_owned_by_user(sk)) { 469 sk->sk_err = err; 470 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ 471 472 tcp_done(sk); 473 } else 474 sk->sk_err_soft = err; 475 goto out; 476 case TCP_LISTEN: 477 break; 478 default: 479 /* check if this ICMP message allows revert of backoff. 480 * (see RFC 6069) 481 */ 482 if (!fastopen && type == ICMPV6_DEST_UNREACH && 483 code == ICMPV6_NOROUTE) 484 tcp_ld_RTO_revert(sk, seq); 485 } 486 487 if (!sock_owned_by_user(sk) && np->recverr) { 488 sk->sk_err = err; 489 sk->sk_error_report(sk); 490 } else 491 sk->sk_err_soft = err; 492 493 out: 494 bh_unlock_sock(sk); 495 sock_put(sk); 496 return 0; 497 } 498 499 500 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, 501 struct flowi *fl, 502 struct request_sock *req, 503 struct tcp_fastopen_cookie *foc, 504 enum tcp_synack_type synack_type) 505 { 506 struct inet_request_sock *ireq = inet_rsk(req); 507 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 508 struct ipv6_txoptions *opt; 509 struct flowi6 *fl6 = &fl->u.ip6; 510 struct sk_buff *skb; 511 int err = -ENOMEM; 512 513 /* First, grab a route. */ 514 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req, 515 IPPROTO_TCP)) == NULL) 516 goto done; 517 518 skb = tcp_make_synack(sk, dst, req, foc, synack_type); 519 520 if (skb) { 521 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, 522 &ireq->ir_v6_rmt_addr); 523 524 fl6->daddr = ireq->ir_v6_rmt_addr; 525 if (np->repflow && ireq->pktopts) 526 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); 527 528 rcu_read_lock(); 529 opt = ireq->ipv6_opt; 530 if (!opt) 531 opt = rcu_dereference(np->opt); 532 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass, 533 sk->sk_priority); 534 rcu_read_unlock(); 535 err = net_xmit_eval(err); 536 } 537 538 done: 539 return err; 540 } 541 542 543 static void tcp_v6_reqsk_destructor(struct request_sock *req) 544 { 545 kfree(inet_rsk(req)->ipv6_opt); 546 kfree_skb(inet_rsk(req)->pktopts); 547 } 548 549 #ifdef CONFIG_TCP_MD5SIG 550 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, 551 const struct in6_addr *addr, 552 int l3index) 553 { 554 return tcp_md5_do_lookup(sk, l3index, 555 (union tcp_md5_addr *)addr, AF_INET6); 556 } 557 558 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk, 559 const struct sock *addr_sk) 560 { 561 int l3index; 562 563 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 564 addr_sk->sk_bound_dev_if); 565 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr, 566 l3index); 567 } 568 569 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname, 570 char __user *optval, int optlen) 571 { 572 struct tcp_md5sig cmd; 573 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; 574 int l3index = 0; 575 u8 prefixlen; 576 577 if (optlen < sizeof(cmd)) 578 return -EINVAL; 579 580 if (copy_from_user(&cmd, optval, sizeof(cmd))) 581 return -EFAULT; 582 583 if (sin6->sin6_family != AF_INET6) 584 return -EINVAL; 585 586 if (optname == TCP_MD5SIG_EXT && 587 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { 588 prefixlen = cmd.tcpm_prefixlen; 589 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) && 590 prefixlen > 32)) 591 return -EINVAL; 592 } else { 593 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128; 594 } 595 596 if (optname == TCP_MD5SIG_EXT && 597 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) { 598 struct net_device *dev; 599 600 rcu_read_lock(); 601 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex); 602 if (dev && netif_is_l3_master(dev)) 603 l3index = dev->ifindex; 604 rcu_read_unlock(); 605 606 /* ok to reference set/not set outside of rcu; 607 * right now device MUST be an L3 master 608 */ 609 if (!dev || !l3index) 610 return -EINVAL; 611 } 612 613 if (!cmd.tcpm_keylen) { 614 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 615 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 616 AF_INET, prefixlen, 617 l3index); 618 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 619 AF_INET6, prefixlen, l3index); 620 } 621 622 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 623 return -EINVAL; 624 625 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 626 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], 627 AF_INET, prefixlen, l3index, 628 cmd.tcpm_key, cmd.tcpm_keylen, 629 GFP_KERNEL); 630 631 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr, 632 AF_INET6, prefixlen, l3index, 633 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL); 634 } 635 636 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp, 637 const struct in6_addr *daddr, 638 const struct in6_addr *saddr, 639 const struct tcphdr *th, int nbytes) 640 { 641 struct tcp6_pseudohdr *bp; 642 struct scatterlist sg; 643 struct tcphdr *_th; 644 645 bp = hp->scratch; 646 /* 1. TCP pseudo-header (RFC2460) */ 647 bp->saddr = *saddr; 648 bp->daddr = *daddr; 649 bp->protocol = cpu_to_be32(IPPROTO_TCP); 650 bp->len = cpu_to_be32(nbytes); 651 652 _th = (struct tcphdr *)(bp + 1); 653 memcpy(_th, th, sizeof(*th)); 654 _th->check = 0; 655 656 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); 657 ahash_request_set_crypt(hp->md5_req, &sg, NULL, 658 sizeof(*bp) + sizeof(*th)); 659 return crypto_ahash_update(hp->md5_req); 660 } 661 662 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 663 const struct in6_addr *daddr, struct in6_addr *saddr, 664 const struct tcphdr *th) 665 { 666 struct tcp_md5sig_pool *hp; 667 struct ahash_request *req; 668 669 hp = tcp_get_md5sig_pool(); 670 if (!hp) 671 goto clear_hash_noput; 672 req = hp->md5_req; 673 674 if (crypto_ahash_init(req)) 675 goto clear_hash; 676 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) 677 goto clear_hash; 678 if (tcp_md5_hash_key(hp, key)) 679 goto clear_hash; 680 ahash_request_set_crypt(req, NULL, md5_hash, 0); 681 if (crypto_ahash_final(req)) 682 goto clear_hash; 683 684 tcp_put_md5sig_pool(); 685 return 0; 686 687 clear_hash: 688 tcp_put_md5sig_pool(); 689 clear_hash_noput: 690 memset(md5_hash, 0, 16); 691 return 1; 692 } 693 694 static int tcp_v6_md5_hash_skb(char *md5_hash, 695 const struct tcp_md5sig_key *key, 696 const struct sock *sk, 697 const struct sk_buff *skb) 698 { 699 const struct in6_addr *saddr, *daddr; 700 struct tcp_md5sig_pool *hp; 701 struct ahash_request *req; 702 const struct tcphdr *th = tcp_hdr(skb); 703 704 if (sk) { /* valid for establish/request sockets */ 705 saddr = &sk->sk_v6_rcv_saddr; 706 daddr = &sk->sk_v6_daddr; 707 } else { 708 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 709 saddr = &ip6h->saddr; 710 daddr = &ip6h->daddr; 711 } 712 713 hp = tcp_get_md5sig_pool(); 714 if (!hp) 715 goto clear_hash_noput; 716 req = hp->md5_req; 717 718 if (crypto_ahash_init(req)) 719 goto clear_hash; 720 721 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len)) 722 goto clear_hash; 723 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 724 goto clear_hash; 725 if (tcp_md5_hash_key(hp, key)) 726 goto clear_hash; 727 ahash_request_set_crypt(req, NULL, md5_hash, 0); 728 if (crypto_ahash_final(req)) 729 goto clear_hash; 730 731 tcp_put_md5sig_pool(); 732 return 0; 733 734 clear_hash: 735 tcp_put_md5sig_pool(); 736 clear_hash_noput: 737 memset(md5_hash, 0, 16); 738 return 1; 739 } 740 741 #endif 742 743 static bool tcp_v6_inbound_md5_hash(const struct sock *sk, 744 const struct sk_buff *skb, 745 int dif, int sdif) 746 { 747 #ifdef CONFIG_TCP_MD5SIG 748 const __u8 *hash_location = NULL; 749 struct tcp_md5sig_key *hash_expected; 750 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 751 const struct tcphdr *th = tcp_hdr(skb); 752 int genhash, l3index; 753 u8 newhash[16]; 754 755 /* sdif set, means packet ingressed via a device 756 * in an L3 domain and dif is set to the l3mdev 757 */ 758 l3index = sdif ? dif : 0; 759 760 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index); 761 hash_location = tcp_parse_md5sig_option(th); 762 763 /* We've parsed the options - do we have a hash? */ 764 if (!hash_expected && !hash_location) 765 return false; 766 767 if (hash_expected && !hash_location) { 768 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 769 return true; 770 } 771 772 if (!hash_expected && hash_location) { 773 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); 774 return true; 775 } 776 777 /* check the signature */ 778 genhash = tcp_v6_md5_hash_skb(newhash, 779 hash_expected, 780 NULL, skb); 781 782 if (genhash || memcmp(hash_location, newhash, 16) != 0) { 783 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE); 784 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n", 785 genhash ? "failed" : "mismatch", 786 &ip6h->saddr, ntohs(th->source), 787 &ip6h->daddr, ntohs(th->dest), l3index); 788 return true; 789 } 790 #endif 791 return false; 792 } 793 794 static void tcp_v6_init_req(struct request_sock *req, 795 const struct sock *sk_listener, 796 struct sk_buff *skb) 797 { 798 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 799 struct inet_request_sock *ireq = inet_rsk(req); 800 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener); 801 802 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; 803 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; 804 805 /* So that link locals have meaning */ 806 if ((!sk_listener->sk_bound_dev_if || l3_slave) && 807 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) 808 ireq->ir_iif = tcp_v6_iif(skb); 809 810 if (!TCP_SKB_CB(skb)->tcp_tw_isn && 811 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) || 812 np->rxopt.bits.rxinfo || 813 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || 814 np->rxopt.bits.rxohlim || np->repflow)) { 815 refcount_inc(&skb->users); 816 ireq->pktopts = skb; 817 } 818 } 819 820 static struct dst_entry *tcp_v6_route_req(const struct sock *sk, 821 struct flowi *fl, 822 const struct request_sock *req) 823 { 824 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP); 825 } 826 827 struct request_sock_ops tcp6_request_sock_ops __read_mostly = { 828 .family = AF_INET6, 829 .obj_size = sizeof(struct tcp6_request_sock), 830 .rtx_syn_ack = tcp_rtx_synack, 831 .send_ack = tcp_v6_reqsk_send_ack, 832 .destructor = tcp_v6_reqsk_destructor, 833 .send_reset = tcp_v6_send_reset, 834 .syn_ack_timeout = tcp_syn_ack_timeout, 835 }; 836 837 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { 838 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 839 sizeof(struct ipv6hdr), 840 #ifdef CONFIG_TCP_MD5SIG 841 .req_md5_lookup = tcp_v6_md5_lookup, 842 .calc_md5_hash = tcp_v6_md5_hash_skb, 843 #endif 844 .init_req = tcp_v6_init_req, 845 #ifdef CONFIG_SYN_COOKIES 846 .cookie_init_seq = cookie_v6_init_sequence, 847 #endif 848 .route_req = tcp_v6_route_req, 849 .init_seq = tcp_v6_init_seq, 850 .init_ts_off = tcp_v6_init_ts_off, 851 .send_synack = tcp_v6_send_synack, 852 }; 853 854 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, 855 u32 ack, u32 win, u32 tsval, u32 tsecr, 856 int oif, struct tcp_md5sig_key *key, int rst, 857 u8 tclass, __be32 label, u32 priority) 858 { 859 const struct tcphdr *th = tcp_hdr(skb); 860 struct tcphdr *t1; 861 struct sk_buff *buff; 862 struct flowi6 fl6; 863 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 864 struct sock *ctl_sk = net->ipv6.tcp_sk; 865 unsigned int tot_len = sizeof(struct tcphdr); 866 struct dst_entry *dst; 867 __be32 *topt; 868 __u32 mark = 0; 869 870 if (tsecr) 871 tot_len += TCPOLEN_TSTAMP_ALIGNED; 872 #ifdef CONFIG_TCP_MD5SIG 873 if (key) 874 tot_len += TCPOLEN_MD5SIG_ALIGNED; 875 #endif 876 877 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len, 878 GFP_ATOMIC); 879 if (!buff) 880 return; 881 882 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len); 883 884 t1 = skb_push(buff, tot_len); 885 skb_reset_transport_header(buff); 886 887 /* Swap the send and the receive. */ 888 memset(t1, 0, sizeof(*t1)); 889 t1->dest = th->source; 890 t1->source = th->dest; 891 t1->doff = tot_len / 4; 892 t1->seq = htonl(seq); 893 t1->ack_seq = htonl(ack); 894 t1->ack = !rst || !th->ack; 895 t1->rst = rst; 896 t1->window = htons(win); 897 898 topt = (__be32 *)(t1 + 1); 899 900 if (tsecr) { 901 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 902 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 903 *topt++ = htonl(tsval); 904 *topt++ = htonl(tsecr); 905 } 906 907 #ifdef CONFIG_TCP_MD5SIG 908 if (key) { 909 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 910 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); 911 tcp_v6_md5_hash_hdr((__u8 *)topt, key, 912 &ipv6_hdr(skb)->saddr, 913 &ipv6_hdr(skb)->daddr, t1); 914 } 915 #endif 916 917 memset(&fl6, 0, sizeof(fl6)); 918 fl6.daddr = ipv6_hdr(skb)->saddr; 919 fl6.saddr = ipv6_hdr(skb)->daddr; 920 fl6.flowlabel = label; 921 922 buff->ip_summed = CHECKSUM_PARTIAL; 923 buff->csum = 0; 924 925 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr); 926 927 fl6.flowi6_proto = IPPROTO_TCP; 928 if (rt6_need_strict(&fl6.daddr) && !oif) 929 fl6.flowi6_oif = tcp_v6_iif(skb); 930 else { 931 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 932 oif = skb->skb_iif; 933 934 fl6.flowi6_oif = oif; 935 } 936 937 if (sk) { 938 if (sk->sk_state == TCP_TIME_WAIT) { 939 mark = inet_twsk(sk)->tw_mark; 940 /* autoflowlabel relies on buff->hash */ 941 skb_set_hash(buff, inet_twsk(sk)->tw_txhash, 942 PKT_HASH_TYPE_L4); 943 } else { 944 mark = sk->sk_mark; 945 } 946 buff->tstamp = tcp_transmit_time(sk); 947 } 948 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark; 949 fl6.fl6_dport = t1->dest; 950 fl6.fl6_sport = t1->source; 951 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); 952 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6)); 953 954 /* Pass a socket to ip6_dst_lookup either it is for RST 955 * Underlying function will use this to retrieve the network 956 * namespace 957 */ 958 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL); 959 if (!IS_ERR(dst)) { 960 skb_dst_set(buff, dst); 961 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass, 962 priority); 963 TCP_INC_STATS(net, TCP_MIB_OUTSEGS); 964 if (rst) 965 TCP_INC_STATS(net, TCP_MIB_OUTRSTS); 966 return; 967 } 968 969 kfree_skb(buff); 970 } 971 972 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) 973 { 974 const struct tcphdr *th = tcp_hdr(skb); 975 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 976 u32 seq = 0, ack_seq = 0; 977 struct tcp_md5sig_key *key = NULL; 978 #ifdef CONFIG_TCP_MD5SIG 979 const __u8 *hash_location = NULL; 980 unsigned char newhash[16]; 981 int genhash; 982 struct sock *sk1 = NULL; 983 #endif 984 __be32 label = 0; 985 u32 priority = 0; 986 struct net *net; 987 int oif = 0; 988 989 if (th->rst) 990 return; 991 992 /* If sk not NULL, it means we did a successful lookup and incoming 993 * route had to be correct. prequeue might have dropped our dst. 994 */ 995 if (!sk && !ipv6_unicast_destination(skb)) 996 return; 997 998 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 999 #ifdef CONFIG_TCP_MD5SIG 1000 rcu_read_lock(); 1001 hash_location = tcp_parse_md5sig_option(th); 1002 if (sk && sk_fullsock(sk)) { 1003 int l3index; 1004 1005 /* sdif set, means packet ingressed via a device 1006 * in an L3 domain and inet_iif is set to it. 1007 */ 1008 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1009 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index); 1010 } else if (hash_location) { 1011 int dif = tcp_v6_iif_l3_slave(skb); 1012 int sdif = tcp_v6_sdif(skb); 1013 int l3index; 1014 1015 /* 1016 * active side is lost. Try to find listening socket through 1017 * source port, and then find md5 key through listening socket. 1018 * we are not loose security here: 1019 * Incoming packet is checked with md5 hash with finding key, 1020 * no RST generated if md5 hash doesn't match. 1021 */ 1022 sk1 = inet6_lookup_listener(net, 1023 &tcp_hashinfo, NULL, 0, 1024 &ipv6h->saddr, 1025 th->source, &ipv6h->daddr, 1026 ntohs(th->source), dif, sdif); 1027 if (!sk1) 1028 goto out; 1029 1030 /* sdif set, means packet ingressed via a device 1031 * in an L3 domain and dif is set to it. 1032 */ 1033 l3index = tcp_v6_sdif(skb) ? dif : 0; 1034 1035 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index); 1036 if (!key) 1037 goto out; 1038 1039 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb); 1040 if (genhash || memcmp(hash_location, newhash, 16) != 0) 1041 goto out; 1042 } 1043 #endif 1044 1045 if (th->ack) 1046 seq = ntohl(th->ack_seq); 1047 else 1048 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - 1049 (th->doff << 2); 1050 1051 if (sk) { 1052 oif = sk->sk_bound_dev_if; 1053 if (sk_fullsock(sk)) { 1054 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1055 1056 trace_tcp_send_reset(sk, skb); 1057 if (np->repflow) 1058 label = ip6_flowlabel(ipv6h); 1059 priority = sk->sk_priority; 1060 } 1061 if (sk->sk_state == TCP_TIME_WAIT) { 1062 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel); 1063 priority = inet_twsk(sk)->tw_priority; 1064 } 1065 } else { 1066 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET) 1067 label = ip6_flowlabel(ipv6h); 1068 } 1069 1070 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 1071 label, priority); 1072 1073 #ifdef CONFIG_TCP_MD5SIG 1074 out: 1075 rcu_read_unlock(); 1076 #endif 1077 } 1078 1079 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, 1080 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, 1081 struct tcp_md5sig_key *key, u8 tclass, 1082 __be32 label, u32 priority) 1083 { 1084 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0, 1085 tclass, label, priority); 1086 } 1087 1088 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) 1089 { 1090 struct inet_timewait_sock *tw = inet_twsk(sk); 1091 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 1092 1093 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 1094 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 1095 tcp_time_stamp_raw() + tcptw->tw_ts_offset, 1096 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw), 1097 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority); 1098 1099 inet_twsk_put(tw); 1100 } 1101 1102 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 1103 struct request_sock *req) 1104 { 1105 int l3index; 1106 1107 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; 1108 1109 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV 1110 * sk->sk_state == TCP_SYN_RECV -> for Fast Open. 1111 */ 1112 /* RFC 7323 2.3 1113 * The window field (SEG.WND) of every outgoing segment, with the 1114 * exception of <SYN> segments, MUST be right-shifted by 1115 * Rcv.Wind.Shift bits: 1116 */ 1117 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ? 1118 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, 1119 tcp_rsk(req)->rcv_nxt, 1120 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, 1121 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off, 1122 req->ts_recent, sk->sk_bound_dev_if, 1123 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index), 1124 0, 0, sk->sk_priority); 1125 } 1126 1127 1128 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) 1129 { 1130 #ifdef CONFIG_SYN_COOKIES 1131 const struct tcphdr *th = tcp_hdr(skb); 1132 1133 if (!th->syn) 1134 sk = cookie_v6_check(sk, skb); 1135 #endif 1136 return sk; 1137 } 1138 1139 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 1140 struct tcphdr *th, u32 *cookie) 1141 { 1142 u16 mss = 0; 1143 #ifdef CONFIG_SYN_COOKIES 1144 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops, 1145 &tcp_request_sock_ipv6_ops, sk, th); 1146 if (mss) { 1147 *cookie = __cookie_v6_init_sequence(iph, th, &mss); 1148 tcp_synq_overflow(sk); 1149 } 1150 #endif 1151 return mss; 1152 } 1153 1154 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 1155 { 1156 if (skb->protocol == htons(ETH_P_IP)) 1157 return tcp_v4_conn_request(sk, skb); 1158 1159 if (!ipv6_unicast_destination(skb)) 1160 goto drop; 1161 1162 return tcp_conn_request(&tcp6_request_sock_ops, 1163 &tcp_request_sock_ipv6_ops, sk, skb); 1164 1165 drop: 1166 tcp_listendrop(sk); 1167 return 0; /* don't send reset */ 1168 } 1169 1170 static void tcp_v6_restore_cb(struct sk_buff *skb) 1171 { 1172 /* We need to move header back to the beginning if xfrm6_policy_check() 1173 * and tcp_v6_fill_cb() are going to be called again. 1174 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there. 1175 */ 1176 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, 1177 sizeof(struct inet6_skb_parm)); 1178 } 1179 1180 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 1181 struct request_sock *req, 1182 struct dst_entry *dst, 1183 struct request_sock *req_unhash, 1184 bool *own_req) 1185 { 1186 struct inet_request_sock *ireq; 1187 struct ipv6_pinfo *newnp; 1188 const struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1189 struct ipv6_txoptions *opt; 1190 struct inet_sock *newinet; 1191 struct tcp_sock *newtp; 1192 struct sock *newsk; 1193 #ifdef CONFIG_TCP_MD5SIG 1194 struct tcp_md5sig_key *key; 1195 int l3index; 1196 #endif 1197 struct flowi6 fl6; 1198 1199 if (skb->protocol == htons(ETH_P_IP)) { 1200 /* 1201 * v6 mapped 1202 */ 1203 1204 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, 1205 req_unhash, own_req); 1206 1207 if (!newsk) 1208 return NULL; 1209 1210 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1211 1212 newinet = inet_sk(newsk); 1213 newnp = tcp_inet6_sk(newsk); 1214 newtp = tcp_sk(newsk); 1215 1216 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1217 1218 newnp->saddr = newsk->sk_v6_rcv_saddr; 1219 1220 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; 1221 if (sk_is_mptcp(newsk)) 1222 mptcpv6_handle_mapped(newsk, true); 1223 newsk->sk_backlog_rcv = tcp_v4_do_rcv; 1224 #ifdef CONFIG_TCP_MD5SIG 1225 newtp->af_specific = &tcp_sock_ipv6_mapped_specific; 1226 #endif 1227 1228 newnp->ipv6_mc_list = NULL; 1229 newnp->ipv6_ac_list = NULL; 1230 newnp->ipv6_fl_list = NULL; 1231 newnp->pktoptions = NULL; 1232 newnp->opt = NULL; 1233 newnp->mcast_oif = inet_iif(skb); 1234 newnp->mcast_hops = ip_hdr(skb)->ttl; 1235 newnp->rcv_flowinfo = 0; 1236 if (np->repflow) 1237 newnp->flow_label = 0; 1238 1239 /* 1240 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 1241 * here, tcp_create_openreq_child now does this for us, see the comment in 1242 * that function for the gory details. -acme 1243 */ 1244 1245 /* It is tricky place. Until this moment IPv4 tcp 1246 worked with IPv6 icsk.icsk_af_ops. 1247 Sync it now. 1248 */ 1249 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 1250 1251 return newsk; 1252 } 1253 1254 ireq = inet_rsk(req); 1255 1256 if (sk_acceptq_is_full(sk)) 1257 goto out_overflow; 1258 1259 if (!dst) { 1260 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP); 1261 if (!dst) 1262 goto out; 1263 } 1264 1265 newsk = tcp_create_openreq_child(sk, req, skb); 1266 if (!newsk) 1267 goto out_nonewsk; 1268 1269 /* 1270 * No need to charge this sock to the relevant IPv6 refcnt debug socks 1271 * count here, tcp_create_openreq_child now does this for us, see the 1272 * comment in that function for the gory details. -acme 1273 */ 1274 1275 newsk->sk_gso_type = SKB_GSO_TCPV6; 1276 ip6_dst_store(newsk, dst, NULL, NULL); 1277 inet6_sk_rx_dst_set(newsk, skb); 1278 1279 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); 1280 1281 newtp = tcp_sk(newsk); 1282 newinet = inet_sk(newsk); 1283 newnp = tcp_inet6_sk(newsk); 1284 1285 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1286 1287 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr; 1288 newnp->saddr = ireq->ir_v6_loc_addr; 1289 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr; 1290 newsk->sk_bound_dev_if = ireq->ir_iif; 1291 1292 /* Now IPv6 options... 1293 1294 First: no IPv4 options. 1295 */ 1296 newinet->inet_opt = NULL; 1297 newnp->ipv6_mc_list = NULL; 1298 newnp->ipv6_ac_list = NULL; 1299 newnp->ipv6_fl_list = NULL; 1300 1301 /* Clone RX bits */ 1302 newnp->rxopt.all = np->rxopt.all; 1303 1304 newnp->pktoptions = NULL; 1305 newnp->opt = NULL; 1306 newnp->mcast_oif = tcp_v6_iif(skb); 1307 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1308 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); 1309 if (np->repflow) 1310 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); 1311 1312 /* Clone native IPv6 options from listening socket (if any) 1313 1314 Yes, keeping reference count would be much more clever, 1315 but we make one more one thing there: reattach optmem 1316 to newsk. 1317 */ 1318 opt = ireq->ipv6_opt; 1319 if (!opt) 1320 opt = rcu_dereference(np->opt); 1321 if (opt) { 1322 opt = ipv6_dup_options(newsk, opt); 1323 RCU_INIT_POINTER(newnp->opt, opt); 1324 } 1325 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1326 if (opt) 1327 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen + 1328 opt->opt_flen; 1329 1330 tcp_ca_openreq_child(newsk, dst); 1331 1332 tcp_sync_mss(newsk, dst_mtu(dst)); 1333 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst)); 1334 1335 tcp_initialize_rcv_mss(newsk); 1336 1337 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 1338 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 1339 1340 #ifdef CONFIG_TCP_MD5SIG 1341 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif); 1342 1343 /* Copy over the MD5 key from the original socket */ 1344 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index); 1345 if (key) { 1346 /* We're using one, so create a matching key 1347 * on the newsk structure. If we fail to get 1348 * memory, then we end up not copying the key 1349 * across. Shucks. 1350 */ 1351 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr, 1352 AF_INET6, 128, l3index, key->key, key->keylen, 1353 sk_gfp_mask(sk, GFP_ATOMIC)); 1354 } 1355 #endif 1356 1357 if (__inet_inherit_port(sk, newsk) < 0) { 1358 inet_csk_prepare_forced_close(newsk); 1359 tcp_done(newsk); 1360 goto out; 1361 } 1362 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); 1363 if (*own_req) { 1364 tcp_move_syn(newtp, req); 1365 1366 /* Clone pktoptions received with SYN, if we own the req */ 1367 if (ireq->pktopts) { 1368 newnp->pktoptions = skb_clone(ireq->pktopts, 1369 sk_gfp_mask(sk, GFP_ATOMIC)); 1370 consume_skb(ireq->pktopts); 1371 ireq->pktopts = NULL; 1372 if (newnp->pktoptions) { 1373 tcp_v6_restore_cb(newnp->pktoptions); 1374 skb_set_owner_r(newnp->pktoptions, newsk); 1375 } 1376 } 1377 } 1378 1379 return newsk; 1380 1381 out_overflow: 1382 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1383 out_nonewsk: 1384 dst_release(dst); 1385 out: 1386 tcp_listendrop(sk); 1387 return NULL; 1388 } 1389 1390 /* The socket must have it's spinlock held when we get 1391 * here, unless it is a TCP_LISTEN socket. 1392 * 1393 * We have a potential double-lock case here, so even when 1394 * doing backlog processing we use the BH locking scheme. 1395 * This is because we cannot sleep with the original spinlock 1396 * held. 1397 */ 1398 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 1399 { 1400 struct ipv6_pinfo *np = tcp_inet6_sk(sk); 1401 struct sk_buff *opt_skb = NULL; 1402 struct tcp_sock *tp; 1403 1404 /* Imagine: socket is IPv6. IPv4 packet arrives, 1405 goes to IPv4 receive handler and backlogged. 1406 From backlog it always goes here. Kerboom... 1407 Fortunately, tcp_rcv_established and rcv_established 1408 handle them correctly, but it is not case with 1409 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK 1410 */ 1411 1412 if (skb->protocol == htons(ETH_P_IP)) 1413 return tcp_v4_do_rcv(sk, skb); 1414 1415 /* 1416 * socket locking is here for SMP purposes as backlog rcv 1417 * is currently called with bh processing disabled. 1418 */ 1419 1420 /* Do Stevens' IPV6_PKTOPTIONS. 1421 1422 Yes, guys, it is the only place in our code, where we 1423 may make it not affecting IPv4. 1424 The rest of code is protocol independent, 1425 and I do not like idea to uglify IPv4. 1426 1427 Actually, all the idea behind IPV6_PKTOPTIONS 1428 looks not very well thought. For now we latch 1429 options, received in the last packet, enqueued 1430 by tcp. Feel free to propose better solution. 1431 --ANK (980728) 1432 */ 1433 if (np->rxopt.all) 1434 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 1435 1436 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1437 struct dst_entry *dst = sk->sk_rx_dst; 1438 1439 sock_rps_save_rxhash(sk, skb); 1440 sk_mark_napi_id(sk, skb); 1441 if (dst) { 1442 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || 1443 dst->ops->check(dst, np->rx_dst_cookie) == NULL) { 1444 dst_release(dst); 1445 sk->sk_rx_dst = NULL; 1446 } 1447 } 1448 1449 tcp_rcv_established(sk, skb); 1450 if (opt_skb) 1451 goto ipv6_pktoptions; 1452 return 0; 1453 } 1454 1455 if (tcp_checksum_complete(skb)) 1456 goto csum_err; 1457 1458 if (sk->sk_state == TCP_LISTEN) { 1459 struct sock *nsk = tcp_v6_cookie_check(sk, skb); 1460 1461 if (!nsk) 1462 goto discard; 1463 1464 if (nsk != sk) { 1465 if (tcp_child_process(sk, nsk, skb)) 1466 goto reset; 1467 if (opt_skb) 1468 __kfree_skb(opt_skb); 1469 return 0; 1470 } 1471 } else 1472 sock_rps_save_rxhash(sk, skb); 1473 1474 if (tcp_rcv_state_process(sk, skb)) 1475 goto reset; 1476 if (opt_skb) 1477 goto ipv6_pktoptions; 1478 return 0; 1479 1480 reset: 1481 tcp_v6_send_reset(sk, skb); 1482 discard: 1483 if (opt_skb) 1484 __kfree_skb(opt_skb); 1485 kfree_skb(skb); 1486 return 0; 1487 csum_err: 1488 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); 1489 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); 1490 goto discard; 1491 1492 1493 ipv6_pktoptions: 1494 /* Do you ask, what is it? 1495 1496 1. skb was enqueued by tcp. 1497 2. skb is added to tail of read queue, rather than out of order. 1498 3. socket is not in passive state. 1499 4. Finally, it really contains options, which user wants to receive. 1500 */ 1501 tp = tcp_sk(sk); 1502 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && 1503 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { 1504 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) 1505 np->mcast_oif = tcp_v6_iif(opt_skb); 1506 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) 1507 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit; 1508 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) 1509 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb)); 1510 if (np->repflow) 1511 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb)); 1512 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) { 1513 skb_set_owner_r(opt_skb, sk); 1514 tcp_v6_restore_cb(opt_skb); 1515 opt_skb = xchg(&np->pktoptions, opt_skb); 1516 } else { 1517 __kfree_skb(opt_skb); 1518 opt_skb = xchg(&np->pktoptions, NULL); 1519 } 1520 } 1521 1522 kfree_skb(opt_skb); 1523 return 0; 1524 } 1525 1526 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, 1527 const struct tcphdr *th) 1528 { 1529 /* This is tricky: we move IP6CB at its correct location into 1530 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because 1531 * _decode_session6() uses IP6CB(). 1532 * barrier() makes sure compiler won't play aliasing games. 1533 */ 1534 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), 1535 sizeof(struct inet6_skb_parm)); 1536 barrier(); 1537 1538 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1539 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1540 skb->len - th->doff*4); 1541 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1542 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1543 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1544 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr); 1545 TCP_SKB_CB(skb)->sacked = 0; 1546 TCP_SKB_CB(skb)->has_rxtstamp = 1547 skb->tstamp || skb_hwtstamps(skb)->hwtstamp; 1548 } 1549 1550 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) 1551 { 1552 struct sk_buff *skb_to_free; 1553 int sdif = inet6_sdif(skb); 1554 int dif = inet6_iif(skb); 1555 const struct tcphdr *th; 1556 const struct ipv6hdr *hdr; 1557 bool refcounted; 1558 struct sock *sk; 1559 int ret; 1560 struct net *net = dev_net(skb->dev); 1561 1562 if (skb->pkt_type != PACKET_HOST) 1563 goto discard_it; 1564 1565 /* 1566 * Count it even if it's bad. 1567 */ 1568 __TCP_INC_STATS(net, TCP_MIB_INSEGS); 1569 1570 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1571 goto discard_it; 1572 1573 th = (const struct tcphdr *)skb->data; 1574 1575 if (unlikely(th->doff < sizeof(struct tcphdr)/4)) 1576 goto bad_packet; 1577 if (!pskb_may_pull(skb, th->doff*4)) 1578 goto discard_it; 1579 1580 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) 1581 goto csum_error; 1582 1583 th = (const struct tcphdr *)skb->data; 1584 hdr = ipv6_hdr(skb); 1585 1586 lookup: 1587 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), 1588 th->source, th->dest, inet6_iif(skb), sdif, 1589 &refcounted); 1590 if (!sk) 1591 goto no_tcp_socket; 1592 1593 process: 1594 if (sk->sk_state == TCP_TIME_WAIT) 1595 goto do_time_wait; 1596 1597 if (sk->sk_state == TCP_NEW_SYN_RECV) { 1598 struct request_sock *req = inet_reqsk(sk); 1599 bool req_stolen = false; 1600 struct sock *nsk; 1601 1602 sk = req->rsk_listener; 1603 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) { 1604 sk_drops_add(sk, skb); 1605 reqsk_put(req); 1606 goto discard_it; 1607 } 1608 if (tcp_checksum_complete(skb)) { 1609 reqsk_put(req); 1610 goto csum_error; 1611 } 1612 if (unlikely(sk->sk_state != TCP_LISTEN)) { 1613 inet_csk_reqsk_queue_drop_and_put(sk, req); 1614 goto lookup; 1615 } 1616 sock_hold(sk); 1617 refcounted = true; 1618 nsk = NULL; 1619 if (!tcp_filter(sk, skb)) { 1620 th = (const struct tcphdr *)skb->data; 1621 hdr = ipv6_hdr(skb); 1622 tcp_v6_fill_cb(skb, hdr, th); 1623 nsk = tcp_check_req(sk, skb, req, false, &req_stolen); 1624 } 1625 if (!nsk) { 1626 reqsk_put(req); 1627 if (req_stolen) { 1628 /* Another cpu got exclusive access to req 1629 * and created a full blown socket. 1630 * Try to feed this packet to this socket 1631 * instead of discarding it. 1632 */ 1633 tcp_v6_restore_cb(skb); 1634 sock_put(sk); 1635 goto lookup; 1636 } 1637 goto discard_and_relse; 1638 } 1639 if (nsk == sk) { 1640 reqsk_put(req); 1641 tcp_v6_restore_cb(skb); 1642 } else if (tcp_child_process(sk, nsk, skb)) { 1643 tcp_v6_send_reset(nsk, skb); 1644 goto discard_and_relse; 1645 } else { 1646 sock_put(sk); 1647 return 0; 1648 } 1649 } 1650 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) { 1651 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); 1652 goto discard_and_relse; 1653 } 1654 1655 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 1656 goto discard_and_relse; 1657 1658 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) 1659 goto discard_and_relse; 1660 1661 if (tcp_filter(sk, skb)) 1662 goto discard_and_relse; 1663 th = (const struct tcphdr *)skb->data; 1664 hdr = ipv6_hdr(skb); 1665 tcp_v6_fill_cb(skb, hdr, th); 1666 1667 skb->dev = NULL; 1668 1669 if (sk->sk_state == TCP_LISTEN) { 1670 ret = tcp_v6_do_rcv(sk, skb); 1671 goto put_and_return; 1672 } 1673 1674 sk_incoming_cpu_update(sk); 1675 1676 bh_lock_sock_nested(sk); 1677 tcp_segs_in(tcp_sk(sk), skb); 1678 ret = 0; 1679 if (!sock_owned_by_user(sk)) { 1680 skb_to_free = sk->sk_rx_skb_cache; 1681 sk->sk_rx_skb_cache = NULL; 1682 ret = tcp_v6_do_rcv(sk, skb); 1683 } else { 1684 if (tcp_add_backlog(sk, skb)) 1685 goto discard_and_relse; 1686 skb_to_free = NULL; 1687 } 1688 bh_unlock_sock(sk); 1689 if (skb_to_free) 1690 __kfree_skb(skb_to_free); 1691 put_and_return: 1692 if (refcounted) 1693 sock_put(sk); 1694 return ret ? -1 : 0; 1695 1696 no_tcp_socket: 1697 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1698 goto discard_it; 1699 1700 tcp_v6_fill_cb(skb, hdr, th); 1701 1702 if (tcp_checksum_complete(skb)) { 1703 csum_error: 1704 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); 1705 bad_packet: 1706 __TCP_INC_STATS(net, TCP_MIB_INERRS); 1707 } else { 1708 tcp_v6_send_reset(NULL, skb); 1709 } 1710 1711 discard_it: 1712 kfree_skb(skb); 1713 return 0; 1714 1715 discard_and_relse: 1716 sk_drops_add(sk, skb); 1717 if (refcounted) 1718 sock_put(sk); 1719 goto discard_it; 1720 1721 do_time_wait: 1722 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1723 inet_twsk_put(inet_twsk(sk)); 1724 goto discard_it; 1725 } 1726 1727 tcp_v6_fill_cb(skb, hdr, th); 1728 1729 if (tcp_checksum_complete(skb)) { 1730 inet_twsk_put(inet_twsk(sk)); 1731 goto csum_error; 1732 } 1733 1734 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1735 case TCP_TW_SYN: 1736 { 1737 struct sock *sk2; 1738 1739 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo, 1740 skb, __tcp_hdrlen(th), 1741 &ipv6_hdr(skb)->saddr, th->source, 1742 &ipv6_hdr(skb)->daddr, 1743 ntohs(th->dest), 1744 tcp_v6_iif_l3_slave(skb), 1745 sdif); 1746 if (sk2) { 1747 struct inet_timewait_sock *tw = inet_twsk(sk); 1748 inet_twsk_deschedule_put(tw); 1749 sk = sk2; 1750 tcp_v6_restore_cb(skb); 1751 refcounted = false; 1752 goto process; 1753 } 1754 } 1755 /* to ACK */ 1756 fallthrough; 1757 case TCP_TW_ACK: 1758 tcp_v6_timewait_ack(sk, skb); 1759 break; 1760 case TCP_TW_RST: 1761 tcp_v6_send_reset(sk, skb); 1762 inet_twsk_deschedule_put(inet_twsk(sk)); 1763 goto discard_it; 1764 case TCP_TW_SUCCESS: 1765 ; 1766 } 1767 goto discard_it; 1768 } 1769 1770 INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb) 1771 { 1772 const struct ipv6hdr *hdr; 1773 const struct tcphdr *th; 1774 struct sock *sk; 1775 1776 if (skb->pkt_type != PACKET_HOST) 1777 return; 1778 1779 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1780 return; 1781 1782 hdr = ipv6_hdr(skb); 1783 th = tcp_hdr(skb); 1784 1785 if (th->doff < sizeof(struct tcphdr) / 4) 1786 return; 1787 1788 /* Note : We use inet6_iif() here, not tcp_v6_iif() */ 1789 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo, 1790 &hdr->saddr, th->source, 1791 &hdr->daddr, ntohs(th->dest), 1792 inet6_iif(skb), inet6_sdif(skb)); 1793 if (sk) { 1794 skb->sk = sk; 1795 skb->destructor = sock_edemux; 1796 if (sk_fullsock(sk)) { 1797 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); 1798 1799 if (dst) 1800 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie); 1801 if (dst && 1802 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) 1803 skb_dst_set_noref(skb, dst); 1804 } 1805 } 1806 } 1807 1808 static struct timewait_sock_ops tcp6_timewait_sock_ops = { 1809 .twsk_obj_size = sizeof(struct tcp6_timewait_sock), 1810 .twsk_unique = tcp_twsk_unique, 1811 .twsk_destructor = tcp_twsk_destructor, 1812 }; 1813 1814 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb) 1815 { 1816 struct ipv6_pinfo *np = inet6_sk(sk); 1817 1818 __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr); 1819 } 1820 1821 const struct inet_connection_sock_af_ops ipv6_specific = { 1822 .queue_xmit = inet6_csk_xmit, 1823 .send_check = tcp_v6_send_check, 1824 .rebuild_header = inet6_sk_rebuild_header, 1825 .sk_rx_dst_set = inet6_sk_rx_dst_set, 1826 .conn_request = tcp_v6_conn_request, 1827 .syn_recv_sock = tcp_v6_syn_recv_sock, 1828 .net_header_len = sizeof(struct ipv6hdr), 1829 .net_frag_header_len = sizeof(struct frag_hdr), 1830 .setsockopt = ipv6_setsockopt, 1831 .getsockopt = ipv6_getsockopt, 1832 .addr2sockaddr = inet6_csk_addr2sockaddr, 1833 .sockaddr_len = sizeof(struct sockaddr_in6), 1834 #ifdef CONFIG_COMPAT 1835 .compat_setsockopt = compat_ipv6_setsockopt, 1836 .compat_getsockopt = compat_ipv6_getsockopt, 1837 #endif 1838 .mtu_reduced = tcp_v6_mtu_reduced, 1839 }; 1840 1841 #ifdef CONFIG_TCP_MD5SIG 1842 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { 1843 .md5_lookup = tcp_v6_md5_lookup, 1844 .calc_md5_hash = tcp_v6_md5_hash_skb, 1845 .md5_parse = tcp_v6_parse_md5_keys, 1846 }; 1847 #endif 1848 1849 /* 1850 * TCP over IPv4 via INET6 API 1851 */ 1852 static const struct inet_connection_sock_af_ops ipv6_mapped = { 1853 .queue_xmit = ip_queue_xmit, 1854 .send_check = tcp_v4_send_check, 1855 .rebuild_header = inet_sk_rebuild_header, 1856 .sk_rx_dst_set = inet_sk_rx_dst_set, 1857 .conn_request = tcp_v6_conn_request, 1858 .syn_recv_sock = tcp_v6_syn_recv_sock, 1859 .net_header_len = sizeof(struct iphdr), 1860 .setsockopt = ipv6_setsockopt, 1861 .getsockopt = ipv6_getsockopt, 1862 .addr2sockaddr = inet6_csk_addr2sockaddr, 1863 .sockaddr_len = sizeof(struct sockaddr_in6), 1864 #ifdef CONFIG_COMPAT 1865 .compat_setsockopt = compat_ipv6_setsockopt, 1866 .compat_getsockopt = compat_ipv6_getsockopt, 1867 #endif 1868 .mtu_reduced = tcp_v4_mtu_reduced, 1869 }; 1870 1871 #ifdef CONFIG_TCP_MD5SIG 1872 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { 1873 .md5_lookup = tcp_v4_md5_lookup, 1874 .calc_md5_hash = tcp_v4_md5_hash_skb, 1875 .md5_parse = tcp_v6_parse_md5_keys, 1876 }; 1877 #endif 1878 1879 /* NOTE: A lot of things set to zero explicitly by call to 1880 * sk_alloc() so need not be done here. 1881 */ 1882 static int tcp_v6_init_sock(struct sock *sk) 1883 { 1884 struct inet_connection_sock *icsk = inet_csk(sk); 1885 1886 tcp_init_sock(sk); 1887 1888 icsk->icsk_af_ops = &ipv6_specific; 1889 1890 #ifdef CONFIG_TCP_MD5SIG 1891 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; 1892 #endif 1893 1894 return 0; 1895 } 1896 1897 static void tcp_v6_destroy_sock(struct sock *sk) 1898 { 1899 tcp_v4_destroy_sock(sk); 1900 inet6_destroy_sock(sk); 1901 } 1902 1903 #ifdef CONFIG_PROC_FS 1904 /* Proc filesystem TCPv6 sock list dumping. */ 1905 static void get_openreq6(struct seq_file *seq, 1906 const struct request_sock *req, int i) 1907 { 1908 long ttd = req->rsk_timer.expires - jiffies; 1909 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr; 1910 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr; 1911 1912 if (ttd < 0) 1913 ttd = 0; 1914 1915 seq_printf(seq, 1916 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1917 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", 1918 i, 1919 src->s6_addr32[0], src->s6_addr32[1], 1920 src->s6_addr32[2], src->s6_addr32[3], 1921 inet_rsk(req)->ir_num, 1922 dest->s6_addr32[0], dest->s6_addr32[1], 1923 dest->s6_addr32[2], dest->s6_addr32[3], 1924 ntohs(inet_rsk(req)->ir_rmt_port), 1925 TCP_SYN_RECV, 1926 0, 0, /* could print option size, but that is af dependent. */ 1927 1, /* timers active (only the expire timer) */ 1928 jiffies_to_clock_t(ttd), 1929 req->num_timeout, 1930 from_kuid_munged(seq_user_ns(seq), 1931 sock_i_uid(req->rsk_listener)), 1932 0, /* non standard timer */ 1933 0, /* open_requests have no inode */ 1934 0, req); 1935 } 1936 1937 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) 1938 { 1939 const struct in6_addr *dest, *src; 1940 __u16 destp, srcp; 1941 int timer_active; 1942 unsigned long timer_expires; 1943 const struct inet_sock *inet = inet_sk(sp); 1944 const struct tcp_sock *tp = tcp_sk(sp); 1945 const struct inet_connection_sock *icsk = inet_csk(sp); 1946 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; 1947 int rx_queue; 1948 int state; 1949 1950 dest = &sp->sk_v6_daddr; 1951 src = &sp->sk_v6_rcv_saddr; 1952 destp = ntohs(inet->inet_dport); 1953 srcp = ntohs(inet->inet_sport); 1954 1955 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 1956 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || 1957 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 1958 timer_active = 1; 1959 timer_expires = icsk->icsk_timeout; 1960 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 1961 timer_active = 4; 1962 timer_expires = icsk->icsk_timeout; 1963 } else if (timer_pending(&sp->sk_timer)) { 1964 timer_active = 2; 1965 timer_expires = sp->sk_timer.expires; 1966 } else { 1967 timer_active = 0; 1968 timer_expires = jiffies; 1969 } 1970 1971 state = inet_sk_state_load(sp); 1972 if (state == TCP_LISTEN) 1973 rx_queue = READ_ONCE(sp->sk_ack_backlog); 1974 else 1975 /* Because we don't lock the socket, 1976 * we might find a transient negative value. 1977 */ 1978 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) - 1979 READ_ONCE(tp->copied_seq), 0); 1980 1981 seq_printf(seq, 1982 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1983 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", 1984 i, 1985 src->s6_addr32[0], src->s6_addr32[1], 1986 src->s6_addr32[2], src->s6_addr32[3], srcp, 1987 dest->s6_addr32[0], dest->s6_addr32[1], 1988 dest->s6_addr32[2], dest->s6_addr32[3], destp, 1989 state, 1990 READ_ONCE(tp->write_seq) - tp->snd_una, 1991 rx_queue, 1992 timer_active, 1993 jiffies_delta_to_clock_t(timer_expires - jiffies), 1994 icsk->icsk_retransmits, 1995 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)), 1996 icsk->icsk_probes_out, 1997 sock_i_ino(sp), 1998 refcount_read(&sp->sk_refcnt), sp, 1999 jiffies_to_clock_t(icsk->icsk_rto), 2000 jiffies_to_clock_t(icsk->icsk_ack.ato), 2001 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp), 2002 tp->snd_cwnd, 2003 state == TCP_LISTEN ? 2004 fastopenq->max_qlen : 2005 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) 2006 ); 2007 } 2008 2009 static void get_timewait6_sock(struct seq_file *seq, 2010 struct inet_timewait_sock *tw, int i) 2011 { 2012 long delta = tw->tw_timer.expires - jiffies; 2013 const struct in6_addr *dest, *src; 2014 __u16 destp, srcp; 2015 2016 dest = &tw->tw_v6_daddr; 2017 src = &tw->tw_v6_rcv_saddr; 2018 destp = ntohs(tw->tw_dport); 2019 srcp = ntohs(tw->tw_sport); 2020 2021 seq_printf(seq, 2022 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 2023 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", 2024 i, 2025 src->s6_addr32[0], src->s6_addr32[1], 2026 src->s6_addr32[2], src->s6_addr32[3], srcp, 2027 dest->s6_addr32[0], dest->s6_addr32[1], 2028 dest->s6_addr32[2], dest->s6_addr32[3], destp, 2029 tw->tw_substate, 0, 0, 2030 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2031 refcount_read(&tw->tw_refcnt), tw); 2032 } 2033 2034 static int tcp6_seq_show(struct seq_file *seq, void *v) 2035 { 2036 struct tcp_iter_state *st; 2037 struct sock *sk = v; 2038 2039 if (v == SEQ_START_TOKEN) { 2040 seq_puts(seq, 2041 " sl " 2042 "local_address " 2043 "remote_address " 2044 "st tx_queue rx_queue tr tm->when retrnsmt" 2045 " uid timeout inode\n"); 2046 goto out; 2047 } 2048 st = seq->private; 2049 2050 if (sk->sk_state == TCP_TIME_WAIT) 2051 get_timewait6_sock(seq, v, st->num); 2052 else if (sk->sk_state == TCP_NEW_SYN_RECV) 2053 get_openreq6(seq, v, st->num); 2054 else 2055 get_tcp6_sock(seq, v, st->num); 2056 out: 2057 return 0; 2058 } 2059 2060 static const struct seq_operations tcp6_seq_ops = { 2061 .show = tcp6_seq_show, 2062 .start = tcp_seq_start, 2063 .next = tcp_seq_next, 2064 .stop = tcp_seq_stop, 2065 }; 2066 2067 static struct tcp_seq_afinfo tcp6_seq_afinfo = { 2068 .family = AF_INET6, 2069 }; 2070 2071 int __net_init tcp6_proc_init(struct net *net) 2072 { 2073 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops, 2074 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo)) 2075 return -ENOMEM; 2076 return 0; 2077 } 2078 2079 void tcp6_proc_exit(struct net *net) 2080 { 2081 remove_proc_entry("tcp6", net->proc_net); 2082 } 2083 #endif 2084 2085 struct proto tcpv6_prot = { 2086 .name = "TCPv6", 2087 .owner = THIS_MODULE, 2088 .close = tcp_close, 2089 .pre_connect = tcp_v6_pre_connect, 2090 .connect = tcp_v6_connect, 2091 .disconnect = tcp_disconnect, 2092 .accept = inet_csk_accept, 2093 .ioctl = tcp_ioctl, 2094 .init = tcp_v6_init_sock, 2095 .destroy = tcp_v6_destroy_sock, 2096 .shutdown = tcp_shutdown, 2097 .setsockopt = tcp_setsockopt, 2098 .getsockopt = tcp_getsockopt, 2099 .keepalive = tcp_set_keepalive, 2100 .recvmsg = tcp_recvmsg, 2101 .sendmsg = tcp_sendmsg, 2102 .sendpage = tcp_sendpage, 2103 .backlog_rcv = tcp_v6_do_rcv, 2104 .release_cb = tcp_release_cb, 2105 .hash = inet6_hash, 2106 .unhash = inet_unhash, 2107 .get_port = inet_csk_get_port, 2108 .enter_memory_pressure = tcp_enter_memory_pressure, 2109 .leave_memory_pressure = tcp_leave_memory_pressure, 2110 .stream_memory_free = tcp_stream_memory_free, 2111 .sockets_allocated = &tcp_sockets_allocated, 2112 .memory_allocated = &tcp_memory_allocated, 2113 .memory_pressure = &tcp_memory_pressure, 2114 .orphan_count = &tcp_orphan_count, 2115 .sysctl_mem = sysctl_tcp_mem, 2116 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), 2117 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), 2118 .max_header = MAX_TCP_HEADER, 2119 .obj_size = sizeof(struct tcp6_sock), 2120 .slab_flags = SLAB_TYPESAFE_BY_RCU, 2121 .twsk_prot = &tcp6_timewait_sock_ops, 2122 .rsk_prot = &tcp6_request_sock_ops, 2123 .h.hashinfo = &tcp_hashinfo, 2124 .no_autobind = true, 2125 #ifdef CONFIG_COMPAT 2126 .compat_setsockopt = compat_tcp_setsockopt, 2127 .compat_getsockopt = compat_tcp_getsockopt, 2128 #endif 2129 .diag_destroy = tcp_abort, 2130 }; 2131 EXPORT_SYMBOL_GPL(tcpv6_prot); 2132 2133 /* thinking of making this const? Don't. 2134 * early_demux can change based on sysctl. 2135 */ 2136 static struct inet6_protocol tcpv6_protocol = { 2137 .early_demux = tcp_v6_early_demux, 2138 .early_demux_handler = tcp_v6_early_demux, 2139 .handler = tcp_v6_rcv, 2140 .err_handler = tcp_v6_err, 2141 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 2142 }; 2143 2144 static struct inet_protosw tcpv6_protosw = { 2145 .type = SOCK_STREAM, 2146 .protocol = IPPROTO_TCP, 2147 .prot = &tcpv6_prot, 2148 .ops = &inet6_stream_ops, 2149 .flags = INET_PROTOSW_PERMANENT | 2150 INET_PROTOSW_ICSK, 2151 }; 2152 2153 static int __net_init tcpv6_net_init(struct net *net) 2154 { 2155 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, 2156 SOCK_RAW, IPPROTO_TCP, net); 2157 } 2158 2159 static void __net_exit tcpv6_net_exit(struct net *net) 2160 { 2161 inet_ctl_sock_destroy(net->ipv6.tcp_sk); 2162 } 2163 2164 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list) 2165 { 2166 inet_twsk_purge(&tcp_hashinfo, AF_INET6); 2167 } 2168 2169 static struct pernet_operations tcpv6_net_ops = { 2170 .init = tcpv6_net_init, 2171 .exit = tcpv6_net_exit, 2172 .exit_batch = tcpv6_net_exit_batch, 2173 }; 2174 2175 int __init tcpv6_init(void) 2176 { 2177 int ret; 2178 2179 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); 2180 if (ret) 2181 goto out; 2182 2183 /* register inet6 protocol */ 2184 ret = inet6_register_protosw(&tcpv6_protosw); 2185 if (ret) 2186 goto out_tcpv6_protocol; 2187 2188 ret = register_pernet_subsys(&tcpv6_net_ops); 2189 if (ret) 2190 goto out_tcpv6_protosw; 2191 2192 ret = mptcpv6_init(); 2193 if (ret) 2194 goto out_tcpv6_pernet_subsys; 2195 2196 out: 2197 return ret; 2198 2199 out_tcpv6_pernet_subsys: 2200 unregister_pernet_subsys(&tcpv6_net_ops); 2201 out_tcpv6_protosw: 2202 inet6_unregister_protosw(&tcpv6_protosw); 2203 out_tcpv6_protocol: 2204 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2205 goto out; 2206 } 2207 2208 void tcpv6_exit(void) 2209 { 2210 unregister_pernet_subsys(&tcpv6_net_ops); 2211 inet6_unregister_protosw(&tcpv6_protosw); 2212 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2213 } 2214