1 /* 2 * TCP over IPv6 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * Based on: 9 * linux/net/ipv4/tcp.c 10 * linux/net/ipv4/tcp_input.c 11 * linux/net/ipv4/tcp_output.c 12 * 13 * Fixes: 14 * Hideaki YOSHIFUJI : sin6_scope_id support 15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 17 * a single port at the same time. 18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file. 19 * 20 * This program is free software; you can redistribute it and/or 21 * modify it under the terms of the GNU General Public License 22 * as published by the Free Software Foundation; either version 23 * 2 of the License, or (at your option) any later version. 24 */ 25 26 #include <linux/bottom_half.h> 27 #include <linux/module.h> 28 #include <linux/errno.h> 29 #include <linux/types.h> 30 #include <linux/socket.h> 31 #include <linux/sockios.h> 32 #include <linux/net.h> 33 #include <linux/jiffies.h> 34 #include <linux/in.h> 35 #include <linux/in6.h> 36 #include <linux/netdevice.h> 37 #include <linux/init.h> 38 #include <linux/jhash.h> 39 #include <linux/ipsec.h> 40 #include <linux/times.h> 41 42 #include <linux/ipv6.h> 43 #include <linux/icmpv6.h> 44 #include <linux/random.h> 45 46 #include <net/tcp.h> 47 #include <net/ndisc.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/inet6_connection_sock.h> 50 #include <net/ipv6.h> 51 #include <net/transp_v6.h> 52 #include <net/addrconf.h> 53 #include <net/ip6_route.h> 54 #include <net/ip6_checksum.h> 55 #include <net/inet_ecn.h> 56 #include <net/protocol.h> 57 #include <net/xfrm.h> 58 #include <net/snmp.h> 59 #include <net/dsfield.h> 60 #include <net/timewait_sock.h> 61 #include <net/netdma.h> 62 #include <net/inet_common.h> 63 64 #include <asm/uaccess.h> 65 66 #include <linux/proc_fs.h> 67 #include <linux/seq_file.h> 68 69 #include <linux/crypto.h> 70 #include <linux/scatterlist.h> 71 72 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb); 73 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, 74 struct request_sock *req); 75 76 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb); 77 78 static struct inet_connection_sock_af_ops ipv6_mapped; 79 static struct inet_connection_sock_af_ops ipv6_specific; 80 #ifdef CONFIG_TCP_MD5SIG 81 static struct tcp_sock_af_ops tcp_sock_ipv6_specific; 82 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific; 83 #else 84 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk, 85 struct in6_addr *addr) 86 { 87 return NULL; 88 } 89 #endif 90 91 static void tcp_v6_hash(struct sock *sk) 92 { 93 if (sk->sk_state != TCP_CLOSE) { 94 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) { 95 tcp_prot.hash(sk); 96 return; 97 } 98 local_bh_disable(); 99 __inet6_hash(sk); 100 local_bh_enable(); 101 } 102 } 103 104 static __inline__ __sum16 tcp_v6_check(int len, 105 struct in6_addr *saddr, 106 struct in6_addr *daddr, 107 __wsum base) 108 { 109 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base); 110 } 111 112 static __u32 tcp_v6_init_sequence(struct sk_buff *skb) 113 { 114 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32, 115 ipv6_hdr(skb)->saddr.s6_addr32, 116 tcp_hdr(skb)->dest, 117 tcp_hdr(skb)->source); 118 } 119 120 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr, 121 int addr_len) 122 { 123 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr; 124 struct inet_sock *inet = inet_sk(sk); 125 struct inet_connection_sock *icsk = inet_csk(sk); 126 struct ipv6_pinfo *np = inet6_sk(sk); 127 struct tcp_sock *tp = tcp_sk(sk); 128 struct in6_addr *saddr = NULL, *final_p = NULL, final; 129 struct flowi fl; 130 struct dst_entry *dst; 131 int addr_type; 132 int err; 133 134 if (addr_len < SIN6_LEN_RFC2133) 135 return -EINVAL; 136 137 if (usin->sin6_family != AF_INET6) 138 return(-EAFNOSUPPORT); 139 140 memset(&fl, 0, sizeof(fl)); 141 142 if (np->sndflow) { 143 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; 144 IP6_ECN_flow_init(fl.fl6_flowlabel); 145 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) { 146 struct ip6_flowlabel *flowlabel; 147 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel); 148 if (flowlabel == NULL) 149 return -EINVAL; 150 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst); 151 fl6_sock_release(flowlabel); 152 } 153 } 154 155 /* 156 * connect() to INADDR_ANY means loopback (BSD'ism). 157 */ 158 159 if(ipv6_addr_any(&usin->sin6_addr)) 160 usin->sin6_addr.s6_addr[15] = 0x1; 161 162 addr_type = ipv6_addr_type(&usin->sin6_addr); 163 164 if(addr_type & IPV6_ADDR_MULTICAST) 165 return -ENETUNREACH; 166 167 if (addr_type&IPV6_ADDR_LINKLOCAL) { 168 if (addr_len >= sizeof(struct sockaddr_in6) && 169 usin->sin6_scope_id) { 170 /* If interface is set while binding, indices 171 * must coincide. 172 */ 173 if (sk->sk_bound_dev_if && 174 sk->sk_bound_dev_if != usin->sin6_scope_id) 175 return -EINVAL; 176 177 sk->sk_bound_dev_if = usin->sin6_scope_id; 178 } 179 180 /* Connect to link-local address requires an interface */ 181 if (!sk->sk_bound_dev_if) 182 return -EINVAL; 183 } 184 185 if (tp->rx_opt.ts_recent_stamp && 186 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) { 187 tp->rx_opt.ts_recent = 0; 188 tp->rx_opt.ts_recent_stamp = 0; 189 tp->write_seq = 0; 190 } 191 192 ipv6_addr_copy(&np->daddr, &usin->sin6_addr); 193 np->flow_label = fl.fl6_flowlabel; 194 195 /* 196 * TCP over IPv4 197 */ 198 199 if (addr_type == IPV6_ADDR_MAPPED) { 200 u32 exthdrlen = icsk->icsk_ext_hdr_len; 201 struct sockaddr_in sin; 202 203 SOCK_DEBUG(sk, "connect: ipv4 mapped\n"); 204 205 if (__ipv6_only_sock(sk)) 206 return -ENETUNREACH; 207 208 sin.sin_family = AF_INET; 209 sin.sin_port = usin->sin6_port; 210 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 211 212 icsk->icsk_af_ops = &ipv6_mapped; 213 sk->sk_backlog_rcv = tcp_v4_do_rcv; 214 #ifdef CONFIG_TCP_MD5SIG 215 tp->af_specific = &tcp_sock_ipv6_mapped_specific; 216 #endif 217 218 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 219 220 if (err) { 221 icsk->icsk_ext_hdr_len = exthdrlen; 222 icsk->icsk_af_ops = &ipv6_specific; 223 sk->sk_backlog_rcv = tcp_v6_do_rcv; 224 #ifdef CONFIG_TCP_MD5SIG 225 tp->af_specific = &tcp_sock_ipv6_specific; 226 #endif 227 goto failure; 228 } else { 229 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF), 230 inet->saddr); 231 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF), 232 inet->rcv_saddr); 233 } 234 235 return err; 236 } 237 238 if (!ipv6_addr_any(&np->rcv_saddr)) 239 saddr = &np->rcv_saddr; 240 241 fl.proto = IPPROTO_TCP; 242 ipv6_addr_copy(&fl.fl6_dst, &np->daddr); 243 ipv6_addr_copy(&fl.fl6_src, 244 (saddr ? saddr : &np->saddr)); 245 fl.oif = sk->sk_bound_dev_if; 246 fl.fl_ip_dport = usin->sin6_port; 247 fl.fl_ip_sport = inet->sport; 248 249 if (np->opt && np->opt->srcrt) { 250 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt; 251 ipv6_addr_copy(&final, &fl.fl6_dst); 252 ipv6_addr_copy(&fl.fl6_dst, rt0->addr); 253 final_p = &final; 254 } 255 256 security_sk_classify_flow(sk, &fl); 257 258 err = ip6_dst_lookup(sk, &dst, &fl); 259 if (err) 260 goto failure; 261 if (final_p) 262 ipv6_addr_copy(&fl.fl6_dst, final_p); 263 264 err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT); 265 if (err < 0) { 266 if (err == -EREMOTE) 267 err = ip6_dst_blackhole(sk, &dst, &fl); 268 if (err < 0) 269 goto failure; 270 } 271 272 if (saddr == NULL) { 273 saddr = &fl.fl6_src; 274 ipv6_addr_copy(&np->rcv_saddr, saddr); 275 } 276 277 /* set the source address */ 278 ipv6_addr_copy(&np->saddr, saddr); 279 inet->rcv_saddr = LOOPBACK4_IPV6; 280 281 sk->sk_gso_type = SKB_GSO_TCPV6; 282 __ip6_dst_store(sk, dst, NULL, NULL); 283 284 icsk->icsk_ext_hdr_len = 0; 285 if (np->opt) 286 icsk->icsk_ext_hdr_len = (np->opt->opt_flen + 287 np->opt->opt_nflen); 288 289 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 290 291 inet->dport = usin->sin6_port; 292 293 tcp_set_state(sk, TCP_SYN_SENT); 294 err = inet6_hash_connect(&tcp_death_row, sk); 295 if (err) 296 goto late_failure; 297 298 if (!tp->write_seq) 299 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32, 300 np->daddr.s6_addr32, 301 inet->sport, 302 inet->dport); 303 304 err = tcp_connect(sk); 305 if (err) 306 goto late_failure; 307 308 return 0; 309 310 late_failure: 311 tcp_set_state(sk, TCP_CLOSE); 312 __sk_dst_reset(sk); 313 failure: 314 inet->dport = 0; 315 sk->sk_route_caps = 0; 316 return err; 317 } 318 319 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 320 int type, int code, int offset, __be32 info) 321 { 322 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data; 323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); 324 struct ipv6_pinfo *np; 325 struct sock *sk; 326 int err; 327 struct tcp_sock *tp; 328 __u32 seq; 329 struct net *net = dev_net(skb->dev); 330 331 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr, 332 th->dest, &hdr->saddr, th->source, skb->dev->ifindex); 333 334 if (sk == NULL) { 335 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 336 ICMP6_MIB_INERRORS); 337 return; 338 } 339 340 if (sk->sk_state == TCP_TIME_WAIT) { 341 inet_twsk_put(inet_twsk(sk)); 342 return; 343 } 344 345 bh_lock_sock(sk); 346 if (sock_owned_by_user(sk)) 347 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); 348 349 if (sk->sk_state == TCP_CLOSE) 350 goto out; 351 352 tp = tcp_sk(sk); 353 seq = ntohl(th->seq); 354 if (sk->sk_state != TCP_LISTEN && 355 !between(seq, tp->snd_una, tp->snd_nxt)) { 356 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 357 goto out; 358 } 359 360 np = inet6_sk(sk); 361 362 if (type == ICMPV6_PKT_TOOBIG) { 363 struct dst_entry *dst = NULL; 364 365 if (sock_owned_by_user(sk)) 366 goto out; 367 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) 368 goto out; 369 370 /* icmp should have updated the destination cache entry */ 371 dst = __sk_dst_check(sk, np->dst_cookie); 372 373 if (dst == NULL) { 374 struct inet_sock *inet = inet_sk(sk); 375 struct flowi fl; 376 377 /* BUGGG_FUTURE: Again, it is not clear how 378 to handle rthdr case. Ignore this complexity 379 for now. 380 */ 381 memset(&fl, 0, sizeof(fl)); 382 fl.proto = IPPROTO_TCP; 383 ipv6_addr_copy(&fl.fl6_dst, &np->daddr); 384 ipv6_addr_copy(&fl.fl6_src, &np->saddr); 385 fl.oif = sk->sk_bound_dev_if; 386 fl.fl_ip_dport = inet->dport; 387 fl.fl_ip_sport = inet->sport; 388 security_skb_classify_flow(skb, &fl); 389 390 if ((err = ip6_dst_lookup(sk, &dst, &fl))) { 391 sk->sk_err_soft = -err; 392 goto out; 393 } 394 395 if ((err = xfrm_lookup(net, &dst, &fl, sk, 0)) < 0) { 396 sk->sk_err_soft = -err; 397 goto out; 398 } 399 400 } else 401 dst_hold(dst); 402 403 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 404 tcp_sync_mss(sk, dst_mtu(dst)); 405 tcp_simple_retransmit(sk); 406 } /* else let the usual retransmit timer handle it */ 407 dst_release(dst); 408 goto out; 409 } 410 411 icmpv6_err_convert(type, code, &err); 412 413 /* Might be for an request_sock */ 414 switch (sk->sk_state) { 415 struct request_sock *req, **prev; 416 case TCP_LISTEN: 417 if (sock_owned_by_user(sk)) 418 goto out; 419 420 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr, 421 &hdr->saddr, inet6_iif(skb)); 422 if (!req) 423 goto out; 424 425 /* ICMPs are not backlogged, hence we cannot get 426 * an established socket here. 427 */ 428 WARN_ON(req->sk != NULL); 429 430 if (seq != tcp_rsk(req)->snt_isn) { 431 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 432 goto out; 433 } 434 435 inet_csk_reqsk_queue_drop(sk, req, prev); 436 goto out; 437 438 case TCP_SYN_SENT: 439 case TCP_SYN_RECV: /* Cannot happen. 440 It can, it SYNs are crossed. --ANK */ 441 if (!sock_owned_by_user(sk)) { 442 sk->sk_err = err; 443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ 444 445 tcp_done(sk); 446 } else 447 sk->sk_err_soft = err; 448 goto out; 449 } 450 451 if (!sock_owned_by_user(sk) && np->recverr) { 452 sk->sk_err = err; 453 sk->sk_error_report(sk); 454 } else 455 sk->sk_err_soft = err; 456 457 out: 458 bh_unlock_sock(sk); 459 sock_put(sk); 460 } 461 462 463 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req) 464 { 465 struct inet6_request_sock *treq = inet6_rsk(req); 466 struct ipv6_pinfo *np = inet6_sk(sk); 467 struct sk_buff * skb; 468 struct ipv6_txoptions *opt = NULL; 469 struct in6_addr * final_p = NULL, final; 470 struct flowi fl; 471 struct dst_entry *dst; 472 int err = -1; 473 474 memset(&fl, 0, sizeof(fl)); 475 fl.proto = IPPROTO_TCP; 476 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr); 477 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr); 478 fl.fl6_flowlabel = 0; 479 fl.oif = treq->iif; 480 fl.fl_ip_dport = inet_rsk(req)->rmt_port; 481 fl.fl_ip_sport = inet_rsk(req)->loc_port; 482 security_req_classify_flow(req, &fl); 483 484 opt = np->opt; 485 if (opt && opt->srcrt) { 486 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt; 487 ipv6_addr_copy(&final, &fl.fl6_dst); 488 ipv6_addr_copy(&fl.fl6_dst, rt0->addr); 489 final_p = &final; 490 } 491 492 err = ip6_dst_lookup(sk, &dst, &fl); 493 if (err) 494 goto done; 495 if (final_p) 496 ipv6_addr_copy(&fl.fl6_dst, final_p); 497 if ((err = xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0) 498 goto done; 499 500 skb = tcp_make_synack(sk, dst, req); 501 if (skb) { 502 struct tcphdr *th = tcp_hdr(skb); 503 504 th->check = tcp_v6_check(skb->len, 505 &treq->loc_addr, &treq->rmt_addr, 506 csum_partial(th, skb->len, skb->csum)); 507 508 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr); 509 err = ip6_xmit(sk, skb, &fl, opt, 0); 510 err = net_xmit_eval(err); 511 } 512 513 done: 514 if (opt && opt != np->opt) 515 sock_kfree_s(sk, opt, opt->tot_len); 516 dst_release(dst); 517 return err; 518 } 519 520 static inline void syn_flood_warning(struct sk_buff *skb) 521 { 522 #ifdef CONFIG_SYN_COOKIES 523 if (sysctl_tcp_syncookies) 524 printk(KERN_INFO 525 "TCPv6: Possible SYN flooding on port %d. " 526 "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest)); 527 else 528 #endif 529 printk(KERN_INFO 530 "TCPv6: Possible SYN flooding on port %d. " 531 "Dropping request.\n", ntohs(tcp_hdr(skb)->dest)); 532 } 533 534 static void tcp_v6_reqsk_destructor(struct request_sock *req) 535 { 536 kfree_skb(inet6_rsk(req)->pktopts); 537 } 538 539 #ifdef CONFIG_TCP_MD5SIG 540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk, 541 struct in6_addr *addr) 542 { 543 struct tcp_sock *tp = tcp_sk(sk); 544 int i; 545 546 BUG_ON(tp == NULL); 547 548 if (!tp->md5sig_info || !tp->md5sig_info->entries6) 549 return NULL; 550 551 for (i = 0; i < tp->md5sig_info->entries6; i++) { 552 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr)) 553 return &tp->md5sig_info->keys6[i].base; 554 } 555 return NULL; 556 } 557 558 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk, 559 struct sock *addr_sk) 560 { 561 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr); 562 } 563 564 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk, 565 struct request_sock *req) 566 { 567 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr); 568 } 569 570 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer, 571 char *newkey, u8 newkeylen) 572 { 573 /* Add key to the list */ 574 struct tcp_md5sig_key *key; 575 struct tcp_sock *tp = tcp_sk(sk); 576 struct tcp6_md5sig_key *keys; 577 578 key = tcp_v6_md5_do_lookup(sk, peer); 579 if (key) { 580 /* modify existing entry - just update that one */ 581 kfree(key->key); 582 key->key = newkey; 583 key->keylen = newkeylen; 584 } else { 585 /* reallocate new list if current one is full. */ 586 if (!tp->md5sig_info) { 587 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC); 588 if (!tp->md5sig_info) { 589 kfree(newkey); 590 return -ENOMEM; 591 } 592 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 593 } 594 if (tcp_alloc_md5sig_pool() == NULL) { 595 kfree(newkey); 596 return -ENOMEM; 597 } 598 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) { 599 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) * 600 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC); 601 602 if (!keys) { 603 tcp_free_md5sig_pool(); 604 kfree(newkey); 605 return -ENOMEM; 606 } 607 608 if (tp->md5sig_info->entries6) 609 memmove(keys, tp->md5sig_info->keys6, 610 (sizeof (tp->md5sig_info->keys6[0]) * 611 tp->md5sig_info->entries6)); 612 613 kfree(tp->md5sig_info->keys6); 614 tp->md5sig_info->keys6 = keys; 615 tp->md5sig_info->alloced6++; 616 } 617 618 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr, 619 peer); 620 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey; 621 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen; 622 623 tp->md5sig_info->entries6++; 624 } 625 return 0; 626 } 627 628 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk, 629 u8 *newkey, __u8 newkeylen) 630 { 631 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr, 632 newkey, newkeylen); 633 } 634 635 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer) 636 { 637 struct tcp_sock *tp = tcp_sk(sk); 638 int i; 639 640 for (i = 0; i < tp->md5sig_info->entries6; i++) { 641 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) { 642 /* Free the key */ 643 kfree(tp->md5sig_info->keys6[i].base.key); 644 tp->md5sig_info->entries6--; 645 646 if (tp->md5sig_info->entries6 == 0) { 647 kfree(tp->md5sig_info->keys6); 648 tp->md5sig_info->keys6 = NULL; 649 tp->md5sig_info->alloced6 = 0; 650 } else { 651 /* shrink the database */ 652 if (tp->md5sig_info->entries6 != i) 653 memmove(&tp->md5sig_info->keys6[i], 654 &tp->md5sig_info->keys6[i+1], 655 (tp->md5sig_info->entries6 - i) 656 * sizeof (tp->md5sig_info->keys6[0])); 657 } 658 tcp_free_md5sig_pool(); 659 return 0; 660 } 661 } 662 return -ENOENT; 663 } 664 665 static void tcp_v6_clear_md5_list (struct sock *sk) 666 { 667 struct tcp_sock *tp = tcp_sk(sk); 668 int i; 669 670 if (tp->md5sig_info->entries6) { 671 for (i = 0; i < tp->md5sig_info->entries6; i++) 672 kfree(tp->md5sig_info->keys6[i].base.key); 673 tp->md5sig_info->entries6 = 0; 674 tcp_free_md5sig_pool(); 675 } 676 677 kfree(tp->md5sig_info->keys6); 678 tp->md5sig_info->keys6 = NULL; 679 tp->md5sig_info->alloced6 = 0; 680 681 if (tp->md5sig_info->entries4) { 682 for (i = 0; i < tp->md5sig_info->entries4; i++) 683 kfree(tp->md5sig_info->keys4[i].base.key); 684 tp->md5sig_info->entries4 = 0; 685 tcp_free_md5sig_pool(); 686 } 687 688 kfree(tp->md5sig_info->keys4); 689 tp->md5sig_info->keys4 = NULL; 690 tp->md5sig_info->alloced4 = 0; 691 } 692 693 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval, 694 int optlen) 695 { 696 struct tcp_md5sig cmd; 697 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; 698 u8 *newkey; 699 700 if (optlen < sizeof(cmd)) 701 return -EINVAL; 702 703 if (copy_from_user(&cmd, optval, sizeof(cmd))) 704 return -EFAULT; 705 706 if (sin6->sin6_family != AF_INET6) 707 return -EINVAL; 708 709 if (!cmd.tcpm_keylen) { 710 if (!tcp_sk(sk)->md5sig_info) 711 return -ENOENT; 712 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) 713 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]); 714 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr); 715 } 716 717 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 718 return -EINVAL; 719 720 if (!tcp_sk(sk)->md5sig_info) { 721 struct tcp_sock *tp = tcp_sk(sk); 722 struct tcp_md5sig_info *p; 723 724 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL); 725 if (!p) 726 return -ENOMEM; 727 728 tp->md5sig_info = p; 729 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 730 } 731 732 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL); 733 if (!newkey) 734 return -ENOMEM; 735 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 736 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3], 737 newkey, cmd.tcpm_keylen); 738 } 739 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen); 740 } 741 742 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, 743 struct in6_addr *daddr, 744 struct in6_addr *saddr, int nbytes) 745 { 746 struct tcp6_pseudohdr *bp; 747 struct scatterlist sg; 748 749 bp = &hp->md5_blk.ip6; 750 /* 1. TCP pseudo-header (RFC2460) */ 751 ipv6_addr_copy(&bp->saddr, saddr); 752 ipv6_addr_copy(&bp->daddr, daddr); 753 bp->protocol = cpu_to_be32(IPPROTO_TCP); 754 bp->len = cpu_to_be32(nbytes); 755 756 sg_init_one(&sg, bp, sizeof(*bp)); 757 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); 758 } 759 760 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key, 761 struct in6_addr *daddr, struct in6_addr *saddr, 762 struct tcphdr *th) 763 { 764 struct tcp_md5sig_pool *hp; 765 struct hash_desc *desc; 766 767 hp = tcp_get_md5sig_pool(); 768 if (!hp) 769 goto clear_hash_noput; 770 desc = &hp->md5_desc; 771 772 if (crypto_hash_init(desc)) 773 goto clear_hash; 774 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) 775 goto clear_hash; 776 if (tcp_md5_hash_header(hp, th)) 777 goto clear_hash; 778 if (tcp_md5_hash_key(hp, key)) 779 goto clear_hash; 780 if (crypto_hash_final(desc, md5_hash)) 781 goto clear_hash; 782 783 tcp_put_md5sig_pool(); 784 return 0; 785 786 clear_hash: 787 tcp_put_md5sig_pool(); 788 clear_hash_noput: 789 memset(md5_hash, 0, 16); 790 return 1; 791 } 792 793 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 794 struct sock *sk, struct request_sock *req, 795 struct sk_buff *skb) 796 { 797 struct in6_addr *saddr, *daddr; 798 struct tcp_md5sig_pool *hp; 799 struct hash_desc *desc; 800 struct tcphdr *th = tcp_hdr(skb); 801 802 if (sk) { 803 saddr = &inet6_sk(sk)->saddr; 804 daddr = &inet6_sk(sk)->daddr; 805 } else if (req) { 806 saddr = &inet6_rsk(req)->loc_addr; 807 daddr = &inet6_rsk(req)->rmt_addr; 808 } else { 809 struct ipv6hdr *ip6h = ipv6_hdr(skb); 810 saddr = &ip6h->saddr; 811 daddr = &ip6h->daddr; 812 } 813 814 hp = tcp_get_md5sig_pool(); 815 if (!hp) 816 goto clear_hash_noput; 817 desc = &hp->md5_desc; 818 819 if (crypto_hash_init(desc)) 820 goto clear_hash; 821 822 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) 823 goto clear_hash; 824 if (tcp_md5_hash_header(hp, th)) 825 goto clear_hash; 826 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 827 goto clear_hash; 828 if (tcp_md5_hash_key(hp, key)) 829 goto clear_hash; 830 if (crypto_hash_final(desc, md5_hash)) 831 goto clear_hash; 832 833 tcp_put_md5sig_pool(); 834 return 0; 835 836 clear_hash: 837 tcp_put_md5sig_pool(); 838 clear_hash_noput: 839 memset(md5_hash, 0, 16); 840 return 1; 841 } 842 843 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb) 844 { 845 __u8 *hash_location = NULL; 846 struct tcp_md5sig_key *hash_expected; 847 struct ipv6hdr *ip6h = ipv6_hdr(skb); 848 struct tcphdr *th = tcp_hdr(skb); 849 int genhash; 850 u8 newhash[16]; 851 852 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr); 853 hash_location = tcp_parse_md5sig_option(th); 854 855 /* We've parsed the options - do we have a hash? */ 856 if (!hash_expected && !hash_location) 857 return 0; 858 859 if (hash_expected && !hash_location) { 860 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 861 return 1; 862 } 863 864 if (!hash_expected && hash_location) { 865 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); 866 return 1; 867 } 868 869 /* check the signature */ 870 genhash = tcp_v6_md5_hash_skb(newhash, 871 hash_expected, 872 NULL, NULL, skb); 873 874 if (genhash || memcmp(hash_location, newhash, 16) != 0) { 875 if (net_ratelimit()) { 876 printk(KERN_INFO "MD5 Hash %s for (%pI6, %u)->(%pI6, %u)\n", 877 genhash ? "failed" : "mismatch", 878 &ip6h->saddr, ntohs(th->source), 879 &ip6h->daddr, ntohs(th->dest)); 880 } 881 return 1; 882 } 883 return 0; 884 } 885 #endif 886 887 struct request_sock_ops tcp6_request_sock_ops __read_mostly = { 888 .family = AF_INET6, 889 .obj_size = sizeof(struct tcp6_request_sock), 890 .rtx_syn_ack = tcp_v6_send_synack, 891 .send_ack = tcp_v6_reqsk_send_ack, 892 .destructor = tcp_v6_reqsk_destructor, 893 .send_reset = tcp_v6_send_reset 894 }; 895 896 #ifdef CONFIG_TCP_MD5SIG 897 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { 898 .md5_lookup = tcp_v6_reqsk_md5_lookup, 899 }; 900 #endif 901 902 static struct timewait_sock_ops tcp6_timewait_sock_ops = { 903 .twsk_obj_size = sizeof(struct tcp6_timewait_sock), 904 .twsk_unique = tcp_twsk_unique, 905 .twsk_destructor= tcp_twsk_destructor, 906 }; 907 908 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb) 909 { 910 struct ipv6_pinfo *np = inet6_sk(sk); 911 struct tcphdr *th = tcp_hdr(skb); 912 913 if (skb->ip_summed == CHECKSUM_PARTIAL) { 914 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0); 915 skb->csum_start = skb_transport_header(skb) - skb->head; 916 skb->csum_offset = offsetof(struct tcphdr, check); 917 } else { 918 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 919 csum_partial(th, th->doff<<2, 920 skb->csum)); 921 } 922 } 923 924 static int tcp_v6_gso_send_check(struct sk_buff *skb) 925 { 926 struct ipv6hdr *ipv6h; 927 struct tcphdr *th; 928 929 if (!pskb_may_pull(skb, sizeof(*th))) 930 return -EINVAL; 931 932 ipv6h = ipv6_hdr(skb); 933 th = tcp_hdr(skb); 934 935 th->check = 0; 936 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len, 937 IPPROTO_TCP, 0); 938 skb->csum_start = skb_transport_header(skb) - skb->head; 939 skb->csum_offset = offsetof(struct tcphdr, check); 940 skb->ip_summed = CHECKSUM_PARTIAL; 941 return 0; 942 } 943 944 struct sk_buff **tcp6_gro_receive(struct sk_buff **head, struct sk_buff *skb) 945 { 946 struct ipv6hdr *iph = ipv6_hdr(skb); 947 948 switch (skb->ip_summed) { 949 case CHECKSUM_COMPLETE: 950 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr, 951 skb->csum)) { 952 skb->ip_summed = CHECKSUM_UNNECESSARY; 953 break; 954 } 955 956 /* fall through */ 957 case CHECKSUM_NONE: 958 NAPI_GRO_CB(skb)->flush = 1; 959 return NULL; 960 } 961 962 return tcp_gro_receive(head, skb); 963 } 964 EXPORT_SYMBOL(tcp6_gro_receive); 965 966 int tcp6_gro_complete(struct sk_buff *skb) 967 { 968 struct ipv6hdr *iph = ipv6_hdr(skb); 969 struct tcphdr *th = tcp_hdr(skb); 970 971 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb), 972 &iph->saddr, &iph->daddr, 0); 973 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; 974 975 return tcp_gro_complete(skb); 976 } 977 EXPORT_SYMBOL(tcp6_gro_complete); 978 979 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win, 980 u32 ts, struct tcp_md5sig_key *key, int rst) 981 { 982 struct tcphdr *th = tcp_hdr(skb), *t1; 983 struct sk_buff *buff; 984 struct flowi fl; 985 struct net *net = dev_net(skb->dst->dev); 986 struct sock *ctl_sk = net->ipv6.tcp_sk; 987 unsigned int tot_len = sizeof(struct tcphdr); 988 __be32 *topt; 989 990 if (ts) 991 tot_len += TCPOLEN_TSTAMP_ALIGNED; 992 #ifdef CONFIG_TCP_MD5SIG 993 if (key) 994 tot_len += TCPOLEN_MD5SIG_ALIGNED; 995 #endif 996 997 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len, 998 GFP_ATOMIC); 999 if (buff == NULL) 1000 return; 1001 1002 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len); 1003 1004 t1 = (struct tcphdr *) skb_push(buff, tot_len); 1005 1006 /* Swap the send and the receive. */ 1007 memset(t1, 0, sizeof(*t1)); 1008 t1->dest = th->source; 1009 t1->source = th->dest; 1010 t1->doff = tot_len / 4; 1011 t1->seq = htonl(seq); 1012 t1->ack_seq = htonl(ack); 1013 t1->ack = !rst || !th->ack; 1014 t1->rst = rst; 1015 t1->window = htons(win); 1016 1017 topt = (__be32 *)(t1 + 1); 1018 1019 if (ts) { 1020 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 1021 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 1022 *topt++ = htonl(tcp_time_stamp); 1023 *topt++ = htonl(ts); 1024 } 1025 1026 #ifdef CONFIG_TCP_MD5SIG 1027 if (key) { 1028 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 1029 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); 1030 tcp_v6_md5_hash_hdr((__u8 *)topt, key, 1031 &ipv6_hdr(skb)->saddr, 1032 &ipv6_hdr(skb)->daddr, t1); 1033 } 1034 #endif 1035 1036 buff->csum = csum_partial(t1, tot_len, 0); 1037 1038 memset(&fl, 0, sizeof(fl)); 1039 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr); 1040 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr); 1041 1042 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst, 1043 tot_len, IPPROTO_TCP, 1044 buff->csum); 1045 1046 fl.proto = IPPROTO_TCP; 1047 fl.oif = inet6_iif(skb); 1048 fl.fl_ip_dport = t1->dest; 1049 fl.fl_ip_sport = t1->source; 1050 security_skb_classify_flow(skb, &fl); 1051 1052 /* Pass a socket to ip6_dst_lookup either it is for RST 1053 * Underlying function will use this to retrieve the network 1054 * namespace 1055 */ 1056 if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) { 1057 if (xfrm_lookup(net, &buff->dst, &fl, NULL, 0) >= 0) { 1058 ip6_xmit(ctl_sk, buff, &fl, NULL, 0); 1059 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); 1060 if (rst) 1061 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); 1062 return; 1063 } 1064 } 1065 1066 kfree_skb(buff); 1067 } 1068 1069 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb) 1070 { 1071 struct tcphdr *th = tcp_hdr(skb); 1072 u32 seq = 0, ack_seq = 0; 1073 struct tcp_md5sig_key *key = NULL; 1074 1075 if (th->rst) 1076 return; 1077 1078 if (!ipv6_unicast_destination(skb)) 1079 return; 1080 1081 #ifdef CONFIG_TCP_MD5SIG 1082 if (sk) 1083 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr); 1084 #endif 1085 1086 if (th->ack) 1087 seq = ntohl(th->ack_seq); 1088 else 1089 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - 1090 (th->doff << 2); 1091 1092 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1); 1093 } 1094 1095 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts, 1096 struct tcp_md5sig_key *key) 1097 { 1098 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0); 1099 } 1100 1101 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) 1102 { 1103 struct inet_timewait_sock *tw = inet_twsk(sk); 1104 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 1105 1106 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 1107 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 1108 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw)); 1109 1110 inet_twsk_put(tw); 1111 } 1112 1113 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, 1114 struct request_sock *req) 1115 { 1116 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent, 1117 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr)); 1118 } 1119 1120 1121 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb) 1122 { 1123 struct request_sock *req, **prev; 1124 const struct tcphdr *th = tcp_hdr(skb); 1125 struct sock *nsk; 1126 1127 /* Find possible connection requests. */ 1128 req = inet6_csk_search_req(sk, &prev, th->source, 1129 &ipv6_hdr(skb)->saddr, 1130 &ipv6_hdr(skb)->daddr, inet6_iif(skb)); 1131 if (req) 1132 return tcp_check_req(sk, skb, req, prev); 1133 1134 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo, 1135 &ipv6_hdr(skb)->saddr, th->source, 1136 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb)); 1137 1138 if (nsk) { 1139 if (nsk->sk_state != TCP_TIME_WAIT) { 1140 bh_lock_sock(nsk); 1141 return nsk; 1142 } 1143 inet_twsk_put(inet_twsk(nsk)); 1144 return NULL; 1145 } 1146 1147 #ifdef CONFIG_SYN_COOKIES 1148 if (!th->rst && !th->syn && th->ack) 1149 sk = cookie_v6_check(sk, skb); 1150 #endif 1151 return sk; 1152 } 1153 1154 /* FIXME: this is substantially similar to the ipv4 code. 1155 * Can some kind of merge be done? -- erics 1156 */ 1157 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 1158 { 1159 struct inet6_request_sock *treq; 1160 struct ipv6_pinfo *np = inet6_sk(sk); 1161 struct tcp_options_received tmp_opt; 1162 struct tcp_sock *tp = tcp_sk(sk); 1163 struct request_sock *req = NULL; 1164 __u32 isn = TCP_SKB_CB(skb)->when; 1165 #ifdef CONFIG_SYN_COOKIES 1166 int want_cookie = 0; 1167 #else 1168 #define want_cookie 0 1169 #endif 1170 1171 if (skb->protocol == htons(ETH_P_IP)) 1172 return tcp_v4_conn_request(sk, skb); 1173 1174 if (!ipv6_unicast_destination(skb)) 1175 goto drop; 1176 1177 if (inet_csk_reqsk_queue_is_full(sk) && !isn) { 1178 if (net_ratelimit()) 1179 syn_flood_warning(skb); 1180 #ifdef CONFIG_SYN_COOKIES 1181 if (sysctl_tcp_syncookies) 1182 want_cookie = 1; 1183 else 1184 #endif 1185 goto drop; 1186 } 1187 1188 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) 1189 goto drop; 1190 1191 req = inet6_reqsk_alloc(&tcp6_request_sock_ops); 1192 if (req == NULL) 1193 goto drop; 1194 1195 #ifdef CONFIG_TCP_MD5SIG 1196 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops; 1197 #endif 1198 1199 tcp_clear_options(&tmp_opt); 1200 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); 1201 tmp_opt.user_mss = tp->rx_opt.user_mss; 1202 1203 tcp_parse_options(skb, &tmp_opt, 0); 1204 1205 if (want_cookie && !tmp_opt.saw_tstamp) 1206 tcp_clear_options(&tmp_opt); 1207 1208 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; 1209 tcp_openreq_init(req, &tmp_opt, skb); 1210 1211 treq = inet6_rsk(req); 1212 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr); 1213 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr); 1214 if (!want_cookie) 1215 TCP_ECN_create_request(req, tcp_hdr(skb)); 1216 1217 if (want_cookie) { 1218 isn = cookie_v6_init_sequence(sk, skb, &req->mss); 1219 req->cookie_ts = tmp_opt.tstamp_ok; 1220 } else if (!isn) { 1221 if (ipv6_opt_accepted(sk, skb) || 1222 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo || 1223 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) { 1224 atomic_inc(&skb->users); 1225 treq->pktopts = skb; 1226 } 1227 treq->iif = sk->sk_bound_dev_if; 1228 1229 /* So that link locals have meaning */ 1230 if (!sk->sk_bound_dev_if && 1231 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL) 1232 treq->iif = inet6_iif(skb); 1233 1234 isn = tcp_v6_init_sequence(skb); 1235 } 1236 1237 tcp_rsk(req)->snt_isn = isn; 1238 1239 security_inet_conn_request(sk, skb, req); 1240 1241 if (tcp_v6_send_synack(sk, req)) 1242 goto drop; 1243 1244 if (!want_cookie) { 1245 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); 1246 return 0; 1247 } 1248 1249 drop: 1250 if (req) 1251 reqsk_free(req); 1252 1253 return 0; /* don't send reset */ 1254 } 1255 1256 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 1257 struct request_sock *req, 1258 struct dst_entry *dst) 1259 { 1260 struct inet6_request_sock *treq; 1261 struct ipv6_pinfo *newnp, *np = inet6_sk(sk); 1262 struct tcp6_sock *newtcp6sk; 1263 struct inet_sock *newinet; 1264 struct tcp_sock *newtp; 1265 struct sock *newsk; 1266 struct ipv6_txoptions *opt; 1267 #ifdef CONFIG_TCP_MD5SIG 1268 struct tcp_md5sig_key *key; 1269 #endif 1270 1271 if (skb->protocol == htons(ETH_P_IP)) { 1272 /* 1273 * v6 mapped 1274 */ 1275 1276 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst); 1277 1278 if (newsk == NULL) 1279 return NULL; 1280 1281 newtcp6sk = (struct tcp6_sock *)newsk; 1282 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6; 1283 1284 newinet = inet_sk(newsk); 1285 newnp = inet6_sk(newsk); 1286 newtp = tcp_sk(newsk); 1287 1288 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1289 1290 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF), 1291 newinet->daddr); 1292 1293 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF), 1294 newinet->saddr); 1295 1296 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr); 1297 1298 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; 1299 newsk->sk_backlog_rcv = tcp_v4_do_rcv; 1300 #ifdef CONFIG_TCP_MD5SIG 1301 newtp->af_specific = &tcp_sock_ipv6_mapped_specific; 1302 #endif 1303 1304 newnp->pktoptions = NULL; 1305 newnp->opt = NULL; 1306 newnp->mcast_oif = inet6_iif(skb); 1307 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1308 1309 /* 1310 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 1311 * here, tcp_create_openreq_child now does this for us, see the comment in 1312 * that function for the gory details. -acme 1313 */ 1314 1315 /* It is tricky place. Until this moment IPv4 tcp 1316 worked with IPv6 icsk.icsk_af_ops. 1317 Sync it now. 1318 */ 1319 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 1320 1321 return newsk; 1322 } 1323 1324 treq = inet6_rsk(req); 1325 opt = np->opt; 1326 1327 if (sk_acceptq_is_full(sk)) 1328 goto out_overflow; 1329 1330 if (dst == NULL) { 1331 struct in6_addr *final_p = NULL, final; 1332 struct flowi fl; 1333 1334 memset(&fl, 0, sizeof(fl)); 1335 fl.proto = IPPROTO_TCP; 1336 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr); 1337 if (opt && opt->srcrt) { 1338 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt; 1339 ipv6_addr_copy(&final, &fl.fl6_dst); 1340 ipv6_addr_copy(&fl.fl6_dst, rt0->addr); 1341 final_p = &final; 1342 } 1343 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr); 1344 fl.oif = sk->sk_bound_dev_if; 1345 fl.fl_ip_dport = inet_rsk(req)->rmt_port; 1346 fl.fl_ip_sport = inet_rsk(req)->loc_port; 1347 security_req_classify_flow(req, &fl); 1348 1349 if (ip6_dst_lookup(sk, &dst, &fl)) 1350 goto out; 1351 1352 if (final_p) 1353 ipv6_addr_copy(&fl.fl6_dst, final_p); 1354 1355 if ((xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0) 1356 goto out; 1357 } 1358 1359 newsk = tcp_create_openreq_child(sk, req, skb); 1360 if (newsk == NULL) 1361 goto out; 1362 1363 /* 1364 * No need to charge this sock to the relevant IPv6 refcnt debug socks 1365 * count here, tcp_create_openreq_child now does this for us, see the 1366 * comment in that function for the gory details. -acme 1367 */ 1368 1369 newsk->sk_gso_type = SKB_GSO_TCPV6; 1370 __ip6_dst_store(newsk, dst, NULL, NULL); 1371 1372 newtcp6sk = (struct tcp6_sock *)newsk; 1373 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6; 1374 1375 newtp = tcp_sk(newsk); 1376 newinet = inet_sk(newsk); 1377 newnp = inet6_sk(newsk); 1378 1379 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 1380 1381 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr); 1382 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr); 1383 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr); 1384 newsk->sk_bound_dev_if = treq->iif; 1385 1386 /* Now IPv6 options... 1387 1388 First: no IPv4 options. 1389 */ 1390 newinet->opt = NULL; 1391 newnp->ipv6_fl_list = NULL; 1392 1393 /* Clone RX bits */ 1394 newnp->rxopt.all = np->rxopt.all; 1395 1396 /* Clone pktoptions received with SYN */ 1397 newnp->pktoptions = NULL; 1398 if (treq->pktopts != NULL) { 1399 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC); 1400 kfree_skb(treq->pktopts); 1401 treq->pktopts = NULL; 1402 if (newnp->pktoptions) 1403 skb_set_owner_r(newnp->pktoptions, newsk); 1404 } 1405 newnp->opt = NULL; 1406 newnp->mcast_oif = inet6_iif(skb); 1407 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 1408 1409 /* Clone native IPv6 options from listening socket (if any) 1410 1411 Yes, keeping reference count would be much more clever, 1412 but we make one more one thing there: reattach optmem 1413 to newsk. 1414 */ 1415 if (opt) { 1416 newnp->opt = ipv6_dup_options(newsk, opt); 1417 if (opt != np->opt) 1418 sock_kfree_s(sk, opt, opt->tot_len); 1419 } 1420 1421 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1422 if (newnp->opt) 1423 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen + 1424 newnp->opt->opt_flen); 1425 1426 tcp_mtup_init(newsk); 1427 tcp_sync_mss(newsk, dst_mtu(dst)); 1428 newtp->advmss = dst_metric(dst, RTAX_ADVMSS); 1429 tcp_initialize_rcv_mss(newsk); 1430 1431 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6; 1432 1433 #ifdef CONFIG_TCP_MD5SIG 1434 /* Copy over the MD5 key from the original socket */ 1435 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) { 1436 /* We're using one, so create a matching key 1437 * on the newsk structure. If we fail to get 1438 * memory, then we end up not copying the key 1439 * across. Shucks. 1440 */ 1441 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC); 1442 if (newkey != NULL) 1443 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr, 1444 newkey, key->keylen); 1445 } 1446 #endif 1447 1448 __inet6_hash(newsk); 1449 __inet_inherit_port(sk, newsk); 1450 1451 return newsk; 1452 1453 out_overflow: 1454 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1455 out: 1456 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 1457 if (opt && opt != np->opt) 1458 sock_kfree_s(sk, opt, opt->tot_len); 1459 dst_release(dst); 1460 return NULL; 1461 } 1462 1463 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb) 1464 { 1465 if (skb->ip_summed == CHECKSUM_COMPLETE) { 1466 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr, 1467 &ipv6_hdr(skb)->daddr, skb->csum)) { 1468 skb->ip_summed = CHECKSUM_UNNECESSARY; 1469 return 0; 1470 } 1471 } 1472 1473 skb->csum = ~csum_unfold(tcp_v6_check(skb->len, 1474 &ipv6_hdr(skb)->saddr, 1475 &ipv6_hdr(skb)->daddr, 0)); 1476 1477 if (skb->len <= 76) { 1478 return __skb_checksum_complete(skb); 1479 } 1480 return 0; 1481 } 1482 1483 /* The socket must have it's spinlock held when we get 1484 * here. 1485 * 1486 * We have a potential double-lock case here, so even when 1487 * doing backlog processing we use the BH locking scheme. 1488 * This is because we cannot sleep with the original spinlock 1489 * held. 1490 */ 1491 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 1492 { 1493 struct ipv6_pinfo *np = inet6_sk(sk); 1494 struct tcp_sock *tp; 1495 struct sk_buff *opt_skb = NULL; 1496 1497 /* Imagine: socket is IPv6. IPv4 packet arrives, 1498 goes to IPv4 receive handler and backlogged. 1499 From backlog it always goes here. Kerboom... 1500 Fortunately, tcp_rcv_established and rcv_established 1501 handle them correctly, but it is not case with 1502 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK 1503 */ 1504 1505 if (skb->protocol == htons(ETH_P_IP)) 1506 return tcp_v4_do_rcv(sk, skb); 1507 1508 #ifdef CONFIG_TCP_MD5SIG 1509 if (tcp_v6_inbound_md5_hash (sk, skb)) 1510 goto discard; 1511 #endif 1512 1513 if (sk_filter(sk, skb)) 1514 goto discard; 1515 1516 /* 1517 * socket locking is here for SMP purposes as backlog rcv 1518 * is currently called with bh processing disabled. 1519 */ 1520 1521 /* Do Stevens' IPV6_PKTOPTIONS. 1522 1523 Yes, guys, it is the only place in our code, where we 1524 may make it not affecting IPv4. 1525 The rest of code is protocol independent, 1526 and I do not like idea to uglify IPv4. 1527 1528 Actually, all the idea behind IPV6_PKTOPTIONS 1529 looks not very well thought. For now we latch 1530 options, received in the last packet, enqueued 1531 by tcp. Feel free to propose better solution. 1532 --ANK (980728) 1533 */ 1534 if (np->rxopt.all) 1535 opt_skb = skb_clone(skb, GFP_ATOMIC); 1536 1537 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1538 TCP_CHECK_TIMER(sk); 1539 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) 1540 goto reset; 1541 TCP_CHECK_TIMER(sk); 1542 if (opt_skb) 1543 goto ipv6_pktoptions; 1544 return 0; 1545 } 1546 1547 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) 1548 goto csum_err; 1549 1550 if (sk->sk_state == TCP_LISTEN) { 1551 struct sock *nsk = tcp_v6_hnd_req(sk, skb); 1552 if (!nsk) 1553 goto discard; 1554 1555 /* 1556 * Queue it on the new socket if the new socket is active, 1557 * otherwise we just shortcircuit this and continue with 1558 * the new socket.. 1559 */ 1560 if(nsk != sk) { 1561 if (tcp_child_process(sk, nsk, skb)) 1562 goto reset; 1563 if (opt_skb) 1564 __kfree_skb(opt_skb); 1565 return 0; 1566 } 1567 } 1568 1569 TCP_CHECK_TIMER(sk); 1570 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) 1571 goto reset; 1572 TCP_CHECK_TIMER(sk); 1573 if (opt_skb) 1574 goto ipv6_pktoptions; 1575 return 0; 1576 1577 reset: 1578 tcp_v6_send_reset(sk, skb); 1579 discard: 1580 if (opt_skb) 1581 __kfree_skb(opt_skb); 1582 kfree_skb(skb); 1583 return 0; 1584 csum_err: 1585 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); 1586 goto discard; 1587 1588 1589 ipv6_pktoptions: 1590 /* Do you ask, what is it? 1591 1592 1. skb was enqueued by tcp. 1593 2. skb is added to tail of read queue, rather than out of order. 1594 3. socket is not in passive state. 1595 4. Finally, it really contains options, which user wants to receive. 1596 */ 1597 tp = tcp_sk(sk); 1598 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && 1599 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { 1600 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) 1601 np->mcast_oif = inet6_iif(opt_skb); 1602 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) 1603 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit; 1604 if (ipv6_opt_accepted(sk, opt_skb)) { 1605 skb_set_owner_r(opt_skb, sk); 1606 opt_skb = xchg(&np->pktoptions, opt_skb); 1607 } else { 1608 __kfree_skb(opt_skb); 1609 opt_skb = xchg(&np->pktoptions, NULL); 1610 } 1611 } 1612 1613 kfree_skb(opt_skb); 1614 return 0; 1615 } 1616 1617 static int tcp_v6_rcv(struct sk_buff *skb) 1618 { 1619 struct tcphdr *th; 1620 struct sock *sk; 1621 int ret; 1622 struct net *net = dev_net(skb->dev); 1623 1624 if (skb->pkt_type != PACKET_HOST) 1625 goto discard_it; 1626 1627 /* 1628 * Count it even if it's bad. 1629 */ 1630 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); 1631 1632 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1633 goto discard_it; 1634 1635 th = tcp_hdr(skb); 1636 1637 if (th->doff < sizeof(struct tcphdr)/4) 1638 goto bad_packet; 1639 if (!pskb_may_pull(skb, th->doff*4)) 1640 goto discard_it; 1641 1642 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb)) 1643 goto bad_packet; 1644 1645 th = tcp_hdr(skb); 1646 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1647 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1648 skb->len - th->doff*4); 1649 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1650 TCP_SKB_CB(skb)->when = 0; 1651 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb)); 1652 TCP_SKB_CB(skb)->sacked = 0; 1653 1654 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); 1655 if (!sk) 1656 goto no_tcp_socket; 1657 1658 process: 1659 if (sk->sk_state == TCP_TIME_WAIT) 1660 goto do_time_wait; 1661 1662 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 1663 goto discard_and_relse; 1664 1665 if (sk_filter(sk, skb)) 1666 goto discard_and_relse; 1667 1668 skb->dev = NULL; 1669 1670 bh_lock_sock_nested(sk); 1671 ret = 0; 1672 if (!sock_owned_by_user(sk)) { 1673 #ifdef CONFIG_NET_DMA 1674 struct tcp_sock *tp = tcp_sk(sk); 1675 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) 1676 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY); 1677 if (tp->ucopy.dma_chan) 1678 ret = tcp_v6_do_rcv(sk, skb); 1679 else 1680 #endif 1681 { 1682 if (!tcp_prequeue(sk, skb)) 1683 ret = tcp_v6_do_rcv(sk, skb); 1684 } 1685 } else 1686 sk_add_backlog(sk, skb); 1687 bh_unlock_sock(sk); 1688 1689 sock_put(sk); 1690 return ret ? -1 : 0; 1691 1692 no_tcp_socket: 1693 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1694 goto discard_it; 1695 1696 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) { 1697 bad_packet: 1698 TCP_INC_STATS_BH(net, TCP_MIB_INERRS); 1699 } else { 1700 tcp_v6_send_reset(NULL, skb); 1701 } 1702 1703 discard_it: 1704 1705 /* 1706 * Discard frame 1707 */ 1708 1709 kfree_skb(skb); 1710 return 0; 1711 1712 discard_and_relse: 1713 sock_put(sk); 1714 goto discard_it; 1715 1716 do_time_wait: 1717 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1718 inet_twsk_put(inet_twsk(sk)); 1719 goto discard_it; 1720 } 1721 1722 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) { 1723 TCP_INC_STATS_BH(net, TCP_MIB_INERRS); 1724 inet_twsk_put(inet_twsk(sk)); 1725 goto discard_it; 1726 } 1727 1728 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1729 case TCP_TW_SYN: 1730 { 1731 struct sock *sk2; 1732 1733 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo, 1734 &ipv6_hdr(skb)->daddr, 1735 ntohs(th->dest), inet6_iif(skb)); 1736 if (sk2 != NULL) { 1737 struct inet_timewait_sock *tw = inet_twsk(sk); 1738 inet_twsk_deschedule(tw, &tcp_death_row); 1739 inet_twsk_put(tw); 1740 sk = sk2; 1741 goto process; 1742 } 1743 /* Fall through to ACK */ 1744 } 1745 case TCP_TW_ACK: 1746 tcp_v6_timewait_ack(sk, skb); 1747 break; 1748 case TCP_TW_RST: 1749 goto no_tcp_socket; 1750 case TCP_TW_SUCCESS:; 1751 } 1752 goto discard_it; 1753 } 1754 1755 static int tcp_v6_remember_stamp(struct sock *sk) 1756 { 1757 /* Alas, not yet... */ 1758 return 0; 1759 } 1760 1761 static struct inet_connection_sock_af_ops ipv6_specific = { 1762 .queue_xmit = inet6_csk_xmit, 1763 .send_check = tcp_v6_send_check, 1764 .rebuild_header = inet6_sk_rebuild_header, 1765 .conn_request = tcp_v6_conn_request, 1766 .syn_recv_sock = tcp_v6_syn_recv_sock, 1767 .remember_stamp = tcp_v6_remember_stamp, 1768 .net_header_len = sizeof(struct ipv6hdr), 1769 .setsockopt = ipv6_setsockopt, 1770 .getsockopt = ipv6_getsockopt, 1771 .addr2sockaddr = inet6_csk_addr2sockaddr, 1772 .sockaddr_len = sizeof(struct sockaddr_in6), 1773 .bind_conflict = inet6_csk_bind_conflict, 1774 #ifdef CONFIG_COMPAT 1775 .compat_setsockopt = compat_ipv6_setsockopt, 1776 .compat_getsockopt = compat_ipv6_getsockopt, 1777 #endif 1778 }; 1779 1780 #ifdef CONFIG_TCP_MD5SIG 1781 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = { 1782 .md5_lookup = tcp_v6_md5_lookup, 1783 .calc_md5_hash = tcp_v6_md5_hash_skb, 1784 .md5_add = tcp_v6_md5_add_func, 1785 .md5_parse = tcp_v6_parse_md5_keys, 1786 }; 1787 #endif 1788 1789 /* 1790 * TCP over IPv4 via INET6 API 1791 */ 1792 1793 static struct inet_connection_sock_af_ops ipv6_mapped = { 1794 .queue_xmit = ip_queue_xmit, 1795 .send_check = tcp_v4_send_check, 1796 .rebuild_header = inet_sk_rebuild_header, 1797 .conn_request = tcp_v6_conn_request, 1798 .syn_recv_sock = tcp_v6_syn_recv_sock, 1799 .remember_stamp = tcp_v4_remember_stamp, 1800 .net_header_len = sizeof(struct iphdr), 1801 .setsockopt = ipv6_setsockopt, 1802 .getsockopt = ipv6_getsockopt, 1803 .addr2sockaddr = inet6_csk_addr2sockaddr, 1804 .sockaddr_len = sizeof(struct sockaddr_in6), 1805 .bind_conflict = inet6_csk_bind_conflict, 1806 #ifdef CONFIG_COMPAT 1807 .compat_setsockopt = compat_ipv6_setsockopt, 1808 .compat_getsockopt = compat_ipv6_getsockopt, 1809 #endif 1810 }; 1811 1812 #ifdef CONFIG_TCP_MD5SIG 1813 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { 1814 .md5_lookup = tcp_v4_md5_lookup, 1815 .calc_md5_hash = tcp_v4_md5_hash_skb, 1816 .md5_add = tcp_v6_md5_add_func, 1817 .md5_parse = tcp_v6_parse_md5_keys, 1818 }; 1819 #endif 1820 1821 /* NOTE: A lot of things set to zero explicitly by call to 1822 * sk_alloc() so need not be done here. 1823 */ 1824 static int tcp_v6_init_sock(struct sock *sk) 1825 { 1826 struct inet_connection_sock *icsk = inet_csk(sk); 1827 struct tcp_sock *tp = tcp_sk(sk); 1828 1829 skb_queue_head_init(&tp->out_of_order_queue); 1830 tcp_init_xmit_timers(sk); 1831 tcp_prequeue_init(tp); 1832 1833 icsk->icsk_rto = TCP_TIMEOUT_INIT; 1834 tp->mdev = TCP_TIMEOUT_INIT; 1835 1836 /* So many TCP implementations out there (incorrectly) count the 1837 * initial SYN frame in their delayed-ACK and congestion control 1838 * algorithms that we must have the following bandaid to talk 1839 * efficiently to them. -DaveM 1840 */ 1841 tp->snd_cwnd = 2; 1842 1843 /* See draft-stevens-tcpca-spec-01 for discussion of the 1844 * initialization of these values. 1845 */ 1846 tp->snd_ssthresh = 0x7fffffff; 1847 tp->snd_cwnd_clamp = ~0; 1848 tp->mss_cache = 536; 1849 1850 tp->reordering = sysctl_tcp_reordering; 1851 1852 sk->sk_state = TCP_CLOSE; 1853 1854 icsk->icsk_af_ops = &ipv6_specific; 1855 icsk->icsk_ca_ops = &tcp_init_congestion_ops; 1856 icsk->icsk_sync_mss = tcp_sync_mss; 1857 sk->sk_write_space = sk_stream_write_space; 1858 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1859 1860 #ifdef CONFIG_TCP_MD5SIG 1861 tp->af_specific = &tcp_sock_ipv6_specific; 1862 #endif 1863 1864 sk->sk_sndbuf = sysctl_tcp_wmem[1]; 1865 sk->sk_rcvbuf = sysctl_tcp_rmem[1]; 1866 1867 local_bh_disable(); 1868 percpu_counter_inc(&tcp_sockets_allocated); 1869 local_bh_enable(); 1870 1871 return 0; 1872 } 1873 1874 static void tcp_v6_destroy_sock(struct sock *sk) 1875 { 1876 #ifdef CONFIG_TCP_MD5SIG 1877 /* Clean up the MD5 key list */ 1878 if (tcp_sk(sk)->md5sig_info) 1879 tcp_v6_clear_md5_list(sk); 1880 #endif 1881 tcp_v4_destroy_sock(sk); 1882 inet6_destroy_sock(sk); 1883 } 1884 1885 #ifdef CONFIG_PROC_FS 1886 /* Proc filesystem TCPv6 sock list dumping. */ 1887 static void get_openreq6(struct seq_file *seq, 1888 struct sock *sk, struct request_sock *req, int i, int uid) 1889 { 1890 int ttd = req->expires - jiffies; 1891 struct in6_addr *src = &inet6_rsk(req)->loc_addr; 1892 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr; 1893 1894 if (ttd < 0) 1895 ttd = 0; 1896 1897 seq_printf(seq, 1898 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1899 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n", 1900 i, 1901 src->s6_addr32[0], src->s6_addr32[1], 1902 src->s6_addr32[2], src->s6_addr32[3], 1903 ntohs(inet_rsk(req)->loc_port), 1904 dest->s6_addr32[0], dest->s6_addr32[1], 1905 dest->s6_addr32[2], dest->s6_addr32[3], 1906 ntohs(inet_rsk(req)->rmt_port), 1907 TCP_SYN_RECV, 1908 0,0, /* could print option size, but that is af dependent. */ 1909 1, /* timers active (only the expire timer) */ 1910 jiffies_to_clock_t(ttd), 1911 req->retrans, 1912 uid, 1913 0, /* non standard timer */ 1914 0, /* open_requests have no inode */ 1915 0, req); 1916 } 1917 1918 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) 1919 { 1920 struct in6_addr *dest, *src; 1921 __u16 destp, srcp; 1922 int timer_active; 1923 unsigned long timer_expires; 1924 struct inet_sock *inet = inet_sk(sp); 1925 struct tcp_sock *tp = tcp_sk(sp); 1926 const struct inet_connection_sock *icsk = inet_csk(sp); 1927 struct ipv6_pinfo *np = inet6_sk(sp); 1928 1929 dest = &np->daddr; 1930 src = &np->rcv_saddr; 1931 destp = ntohs(inet->dport); 1932 srcp = ntohs(inet->sport); 1933 1934 if (icsk->icsk_pending == ICSK_TIME_RETRANS) { 1935 timer_active = 1; 1936 timer_expires = icsk->icsk_timeout; 1937 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 1938 timer_active = 4; 1939 timer_expires = icsk->icsk_timeout; 1940 } else if (timer_pending(&sp->sk_timer)) { 1941 timer_active = 2; 1942 timer_expires = sp->sk_timer.expires; 1943 } else { 1944 timer_active = 0; 1945 timer_expires = jiffies; 1946 } 1947 1948 seq_printf(seq, 1949 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1950 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n", 1951 i, 1952 src->s6_addr32[0], src->s6_addr32[1], 1953 src->s6_addr32[2], src->s6_addr32[3], srcp, 1954 dest->s6_addr32[0], dest->s6_addr32[1], 1955 dest->s6_addr32[2], dest->s6_addr32[3], destp, 1956 sp->sk_state, 1957 tp->write_seq-tp->snd_una, 1958 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq), 1959 timer_active, 1960 jiffies_to_clock_t(timer_expires - jiffies), 1961 icsk->icsk_retransmits, 1962 sock_i_uid(sp), 1963 icsk->icsk_probes_out, 1964 sock_i_ino(sp), 1965 atomic_read(&sp->sk_refcnt), sp, 1966 jiffies_to_clock_t(icsk->icsk_rto), 1967 jiffies_to_clock_t(icsk->icsk_ack.ato), 1968 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong, 1969 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh 1970 ); 1971 } 1972 1973 static void get_timewait6_sock(struct seq_file *seq, 1974 struct inet_timewait_sock *tw, int i) 1975 { 1976 struct in6_addr *dest, *src; 1977 __u16 destp, srcp; 1978 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw); 1979 int ttd = tw->tw_ttd - jiffies; 1980 1981 if (ttd < 0) 1982 ttd = 0; 1983 1984 dest = &tw6->tw_v6_daddr; 1985 src = &tw6->tw_v6_rcv_saddr; 1986 destp = ntohs(tw->tw_dport); 1987 srcp = ntohs(tw->tw_sport); 1988 1989 seq_printf(seq, 1990 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1991 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n", 1992 i, 1993 src->s6_addr32[0], src->s6_addr32[1], 1994 src->s6_addr32[2], src->s6_addr32[3], srcp, 1995 dest->s6_addr32[0], dest->s6_addr32[1], 1996 dest->s6_addr32[2], dest->s6_addr32[3], destp, 1997 tw->tw_substate, 0, 0, 1998 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, 1999 atomic_read(&tw->tw_refcnt), tw); 2000 } 2001 2002 static int tcp6_seq_show(struct seq_file *seq, void *v) 2003 { 2004 struct tcp_iter_state *st; 2005 2006 if (v == SEQ_START_TOKEN) { 2007 seq_puts(seq, 2008 " sl " 2009 "local_address " 2010 "remote_address " 2011 "st tx_queue rx_queue tr tm->when retrnsmt" 2012 " uid timeout inode\n"); 2013 goto out; 2014 } 2015 st = seq->private; 2016 2017 switch (st->state) { 2018 case TCP_SEQ_STATE_LISTENING: 2019 case TCP_SEQ_STATE_ESTABLISHED: 2020 get_tcp6_sock(seq, v, st->num); 2021 break; 2022 case TCP_SEQ_STATE_OPENREQ: 2023 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid); 2024 break; 2025 case TCP_SEQ_STATE_TIME_WAIT: 2026 get_timewait6_sock(seq, v, st->num); 2027 break; 2028 } 2029 out: 2030 return 0; 2031 } 2032 2033 static struct tcp_seq_afinfo tcp6_seq_afinfo = { 2034 .name = "tcp6", 2035 .family = AF_INET6, 2036 .seq_fops = { 2037 .owner = THIS_MODULE, 2038 }, 2039 .seq_ops = { 2040 .show = tcp6_seq_show, 2041 }, 2042 }; 2043 2044 int tcp6_proc_init(struct net *net) 2045 { 2046 return tcp_proc_register(net, &tcp6_seq_afinfo); 2047 } 2048 2049 void tcp6_proc_exit(struct net *net) 2050 { 2051 tcp_proc_unregister(net, &tcp6_seq_afinfo); 2052 } 2053 #endif 2054 2055 struct proto tcpv6_prot = { 2056 .name = "TCPv6", 2057 .owner = THIS_MODULE, 2058 .close = tcp_close, 2059 .connect = tcp_v6_connect, 2060 .disconnect = tcp_disconnect, 2061 .accept = inet_csk_accept, 2062 .ioctl = tcp_ioctl, 2063 .init = tcp_v6_init_sock, 2064 .destroy = tcp_v6_destroy_sock, 2065 .shutdown = tcp_shutdown, 2066 .setsockopt = tcp_setsockopt, 2067 .getsockopt = tcp_getsockopt, 2068 .recvmsg = tcp_recvmsg, 2069 .backlog_rcv = tcp_v6_do_rcv, 2070 .hash = tcp_v6_hash, 2071 .unhash = inet_unhash, 2072 .get_port = inet_csk_get_port, 2073 .enter_memory_pressure = tcp_enter_memory_pressure, 2074 .sockets_allocated = &tcp_sockets_allocated, 2075 .memory_allocated = &tcp_memory_allocated, 2076 .memory_pressure = &tcp_memory_pressure, 2077 .orphan_count = &tcp_orphan_count, 2078 .sysctl_mem = sysctl_tcp_mem, 2079 .sysctl_wmem = sysctl_tcp_wmem, 2080 .sysctl_rmem = sysctl_tcp_rmem, 2081 .max_header = MAX_TCP_HEADER, 2082 .obj_size = sizeof(struct tcp6_sock), 2083 .slab_flags = SLAB_DESTROY_BY_RCU, 2084 .twsk_prot = &tcp6_timewait_sock_ops, 2085 .rsk_prot = &tcp6_request_sock_ops, 2086 .h.hashinfo = &tcp_hashinfo, 2087 #ifdef CONFIG_COMPAT 2088 .compat_setsockopt = compat_tcp_setsockopt, 2089 .compat_getsockopt = compat_tcp_getsockopt, 2090 #endif 2091 }; 2092 2093 static struct inet6_protocol tcpv6_protocol = { 2094 .handler = tcp_v6_rcv, 2095 .err_handler = tcp_v6_err, 2096 .gso_send_check = tcp_v6_gso_send_check, 2097 .gso_segment = tcp_tso_segment, 2098 .gro_receive = tcp6_gro_receive, 2099 .gro_complete = tcp6_gro_complete, 2100 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 2101 }; 2102 2103 static struct inet_protosw tcpv6_protosw = { 2104 .type = SOCK_STREAM, 2105 .protocol = IPPROTO_TCP, 2106 .prot = &tcpv6_prot, 2107 .ops = &inet6_stream_ops, 2108 .capability = -1, 2109 .no_check = 0, 2110 .flags = INET_PROTOSW_PERMANENT | 2111 INET_PROTOSW_ICSK, 2112 }; 2113 2114 static int tcpv6_net_init(struct net *net) 2115 { 2116 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, 2117 SOCK_RAW, IPPROTO_TCP, net); 2118 } 2119 2120 static void tcpv6_net_exit(struct net *net) 2121 { 2122 inet_ctl_sock_destroy(net->ipv6.tcp_sk); 2123 inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET6); 2124 } 2125 2126 static struct pernet_operations tcpv6_net_ops = { 2127 .init = tcpv6_net_init, 2128 .exit = tcpv6_net_exit, 2129 }; 2130 2131 int __init tcpv6_init(void) 2132 { 2133 int ret; 2134 2135 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); 2136 if (ret) 2137 goto out; 2138 2139 /* register inet6 protocol */ 2140 ret = inet6_register_protosw(&tcpv6_protosw); 2141 if (ret) 2142 goto out_tcpv6_protocol; 2143 2144 ret = register_pernet_subsys(&tcpv6_net_ops); 2145 if (ret) 2146 goto out_tcpv6_protosw; 2147 out: 2148 return ret; 2149 2150 out_tcpv6_protocol: 2151 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2152 out_tcpv6_protosw: 2153 inet6_unregister_protosw(&tcpv6_protosw); 2154 goto out; 2155 } 2156 2157 void tcpv6_exit(void) 2158 { 2159 unregister_pernet_subsys(&tcpv6_net_ops); 2160 inet6_unregister_protosw(&tcpv6_protosw); 2161 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); 2162 } 2163