1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Implementation of the Transmission Control Protocol(TCP). 7 * 8 * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $ 9 * 10 * IPv4 specific functions 11 * 12 * 13 * code split from: 14 * linux/ipv4/tcp.c 15 * linux/ipv4/tcp_input.c 16 * linux/ipv4/tcp_output.c 17 * 18 * See tcp.c for author information 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 /* 27 * Changes: 28 * David S. Miller : New socket lookup architecture. 29 * This code is dedicated to John Dyson. 30 * David S. Miller : Change semantics of established hash, 31 * half is devoted to TIME_WAIT sockets 32 * and the rest go in the other half. 33 * Andi Kleen : Add support for syncookies and fixed 34 * some bugs: ip options weren't passed to 35 * the TCP layer, missed a check for an 36 * ACK bit. 37 * Andi Kleen : Implemented fast path mtu discovery. 38 * Fixed many serious bugs in the 39 * request_sock handling and moved 40 * most of it into the af independent code. 41 * Added tail drop and some other bugfixes. 42 * Added new listen semantics. 43 * Mike McLagan : Routing by source 44 * Juan Jose Ciarlante: ip_dynaddr bits 45 * Andi Kleen: various fixes. 46 * Vitaly E. Lavrov : Transparent proxy revived after year 47 * coma. 48 * Andi Kleen : Fix new listen. 49 * Andi Kleen : Fix accept error reporting. 50 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 51 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 52 * a single port at the same time. 53 */ 54 55 56 #include <linux/types.h> 57 #include <linux/fcntl.h> 58 #include <linux/module.h> 59 #include <linux/random.h> 60 #include <linux/cache.h> 61 #include <linux/jhash.h> 62 #include <linux/init.h> 63 #include <linux/times.h> 64 65 #include <net/icmp.h> 66 #include <net/inet_hashtables.h> 67 #include <net/tcp.h> 68 #include <net/transp_v6.h> 69 #include <net/ipv6.h> 70 #include <net/inet_common.h> 71 #include <net/timewait_sock.h> 72 #include <net/xfrm.h> 73 #include <net/netdma.h> 74 75 #include <linux/inet.h> 76 #include <linux/ipv6.h> 77 #include <linux/stddef.h> 78 #include <linux/proc_fs.h> 79 #include <linux/seq_file.h> 80 81 int sysctl_tcp_tw_reuse; 82 int sysctl_tcp_low_latency; 83 84 /* Check TCP sequence numbers in ICMP packets. */ 85 #define ICMP_MIN_LENGTH 8 86 87 /* Socket used for sending RSTs */ 88 static struct socket *tcp_socket; 89 90 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb); 91 92 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = { 93 .lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock), 94 .lhash_users = ATOMIC_INIT(0), 95 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait), 96 }; 97 98 static int tcp_v4_get_port(struct sock *sk, unsigned short snum) 99 { 100 return inet_csk_get_port(&tcp_hashinfo, sk, snum, 101 inet_csk_bind_conflict); 102 } 103 104 static void tcp_v4_hash(struct sock *sk) 105 { 106 inet_hash(&tcp_hashinfo, sk); 107 } 108 109 void tcp_unhash(struct sock *sk) 110 { 111 inet_unhash(&tcp_hashinfo, sk); 112 } 113 114 static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb) 115 { 116 return secure_tcp_sequence_number(skb->nh.iph->daddr, 117 skb->nh.iph->saddr, 118 skb->h.th->dest, 119 skb->h.th->source); 120 } 121 122 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) 123 { 124 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); 125 struct tcp_sock *tp = tcp_sk(sk); 126 127 /* With PAWS, it is safe from the viewpoint 128 of data integrity. Even without PAWS it is safe provided sequence 129 spaces do not overlap i.e. at data rates <= 80Mbit/sec. 130 131 Actually, the idea is close to VJ's one, only timestamp cache is 132 held not per host, but per port pair and TW bucket is used as state 133 holder. 134 135 If TW bucket has been already destroyed we fall back to VJ's scheme 136 and use initial timestamp retrieved from peer table. 137 */ 138 if (tcptw->tw_ts_recent_stamp && 139 (twp == NULL || (sysctl_tcp_tw_reuse && 140 xtime.tv_sec - tcptw->tw_ts_recent_stamp > 1))) { 141 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; 142 if (tp->write_seq == 0) 143 tp->write_seq = 1; 144 tp->rx_opt.ts_recent = tcptw->tw_ts_recent; 145 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; 146 sock_hold(sktw); 147 return 1; 148 } 149 150 return 0; 151 } 152 153 EXPORT_SYMBOL_GPL(tcp_twsk_unique); 154 155 /* This will initiate an outgoing connection. */ 156 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 157 { 158 struct inet_sock *inet = inet_sk(sk); 159 struct tcp_sock *tp = tcp_sk(sk); 160 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; 161 struct rtable *rt; 162 u32 daddr, nexthop; 163 int tmp; 164 int err; 165 166 if (addr_len < sizeof(struct sockaddr_in)) 167 return -EINVAL; 168 169 if (usin->sin_family != AF_INET) 170 return -EAFNOSUPPORT; 171 172 nexthop = daddr = usin->sin_addr.s_addr; 173 if (inet->opt && inet->opt->srr) { 174 if (!daddr) 175 return -EINVAL; 176 nexthop = inet->opt->faddr; 177 } 178 179 tmp = ip_route_connect(&rt, nexthop, inet->saddr, 180 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, 181 IPPROTO_TCP, 182 inet->sport, usin->sin_port, sk); 183 if (tmp < 0) 184 return tmp; 185 186 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { 187 ip_rt_put(rt); 188 return -ENETUNREACH; 189 } 190 191 if (!inet->opt || !inet->opt->srr) 192 daddr = rt->rt_dst; 193 194 if (!inet->saddr) 195 inet->saddr = rt->rt_src; 196 inet->rcv_saddr = inet->saddr; 197 198 if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { 199 /* Reset inherited state */ 200 tp->rx_opt.ts_recent = 0; 201 tp->rx_opt.ts_recent_stamp = 0; 202 tp->write_seq = 0; 203 } 204 205 if (tcp_death_row.sysctl_tw_recycle && 206 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { 207 struct inet_peer *peer = rt_get_peer(rt); 208 209 /* VJ's idea. We save last timestamp seen from 210 * the destination in peer table, when entering state TIME-WAIT 211 * and initialize rx_opt.ts_recent from it, when trying new connection. 212 */ 213 214 if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) { 215 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; 216 tp->rx_opt.ts_recent = peer->tcp_ts; 217 } 218 } 219 220 inet->dport = usin->sin_port; 221 inet->daddr = daddr; 222 223 inet_csk(sk)->icsk_ext_hdr_len = 0; 224 if (inet->opt) 225 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen; 226 227 tp->rx_opt.mss_clamp = 536; 228 229 /* Socket identity is still unknown (sport may be zero). 230 * However we set state to SYN-SENT and not releasing socket 231 * lock select source port, enter ourselves into the hash tables and 232 * complete initialization after this. 233 */ 234 tcp_set_state(sk, TCP_SYN_SENT); 235 err = inet_hash_connect(&tcp_death_row, sk); 236 if (err) 237 goto failure; 238 239 err = ip_route_newports(&rt, IPPROTO_TCP, inet->sport, inet->dport, sk); 240 if (err) 241 goto failure; 242 243 /* OK, now commit destination to socket. */ 244 sk->sk_gso_type = SKB_GSO_TCPV4; 245 sk_setup_caps(sk, &rt->u.dst); 246 247 if (!tp->write_seq) 248 tp->write_seq = secure_tcp_sequence_number(inet->saddr, 249 inet->daddr, 250 inet->sport, 251 usin->sin_port); 252 253 inet->id = tp->write_seq ^ jiffies; 254 255 err = tcp_connect(sk); 256 rt = NULL; 257 if (err) 258 goto failure; 259 260 return 0; 261 262 failure: 263 /* This unhashes the socket and releases the local port, if necessary. */ 264 tcp_set_state(sk, TCP_CLOSE); 265 ip_rt_put(rt); 266 sk->sk_route_caps = 0; 267 inet->dport = 0; 268 return err; 269 } 270 271 /* 272 * This routine does path mtu discovery as defined in RFC1191. 273 */ 274 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu) 275 { 276 struct dst_entry *dst; 277 struct inet_sock *inet = inet_sk(sk); 278 279 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs 280 * send out by Linux are always <576bytes so they should go through 281 * unfragmented). 282 */ 283 if (sk->sk_state == TCP_LISTEN) 284 return; 285 286 /* We don't check in the destentry if pmtu discovery is forbidden 287 * on this route. We just assume that no packet_to_big packets 288 * are send back when pmtu discovery is not active. 289 * There is a small race when the user changes this flag in the 290 * route, but I think that's acceptable. 291 */ 292 if ((dst = __sk_dst_check(sk, 0)) == NULL) 293 return; 294 295 dst->ops->update_pmtu(dst, mtu); 296 297 /* Something is about to be wrong... Remember soft error 298 * for the case, if this connection will not able to recover. 299 */ 300 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) 301 sk->sk_err_soft = EMSGSIZE; 302 303 mtu = dst_mtu(dst); 304 305 if (inet->pmtudisc != IP_PMTUDISC_DONT && 306 inet_csk(sk)->icsk_pmtu_cookie > mtu) { 307 tcp_sync_mss(sk, mtu); 308 309 /* Resend the TCP packet because it's 310 * clear that the old packet has been 311 * dropped. This is the new "fast" path mtu 312 * discovery. 313 */ 314 tcp_simple_retransmit(sk); 315 } /* else let the usual retransmit timer handle it */ 316 } 317 318 /* 319 * This routine is called by the ICMP module when it gets some 320 * sort of error condition. If err < 0 then the socket should 321 * be closed and the error returned to the user. If err > 0 322 * it's just the icmp type << 8 | icmp code. After adjustment 323 * header points to the first 8 bytes of the tcp header. We need 324 * to find the appropriate port. 325 * 326 * The locking strategy used here is very "optimistic". When 327 * someone else accesses the socket the ICMP is just dropped 328 * and for some paths there is no check at all. 329 * A more general error queue to queue errors for later handling 330 * is probably better. 331 * 332 */ 333 334 void tcp_v4_err(struct sk_buff *skb, u32 info) 335 { 336 struct iphdr *iph = (struct iphdr *)skb->data; 337 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); 338 struct tcp_sock *tp; 339 struct inet_sock *inet; 340 int type = skb->h.icmph->type; 341 int code = skb->h.icmph->code; 342 struct sock *sk; 343 __u32 seq; 344 int err; 345 346 if (skb->len < (iph->ihl << 2) + 8) { 347 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 348 return; 349 } 350 351 sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr, 352 th->source, inet_iif(skb)); 353 if (!sk) { 354 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 355 return; 356 } 357 if (sk->sk_state == TCP_TIME_WAIT) { 358 inet_twsk_put((struct inet_timewait_sock *)sk); 359 return; 360 } 361 362 bh_lock_sock(sk); 363 /* If too many ICMPs get dropped on busy 364 * servers this needs to be solved differently. 365 */ 366 if (sock_owned_by_user(sk)) 367 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); 368 369 if (sk->sk_state == TCP_CLOSE) 370 goto out; 371 372 tp = tcp_sk(sk); 373 seq = ntohl(th->seq); 374 if (sk->sk_state != TCP_LISTEN && 375 !between(seq, tp->snd_una, tp->snd_nxt)) { 376 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS); 377 goto out; 378 } 379 380 switch (type) { 381 case ICMP_SOURCE_QUENCH: 382 /* Just silently ignore these. */ 383 goto out; 384 case ICMP_PARAMETERPROB: 385 err = EPROTO; 386 break; 387 case ICMP_DEST_UNREACH: 388 if (code > NR_ICMP_UNREACH) 389 goto out; 390 391 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ 392 if (!sock_owned_by_user(sk)) 393 do_pmtu_discovery(sk, iph, info); 394 goto out; 395 } 396 397 err = icmp_err_convert[code].errno; 398 break; 399 case ICMP_TIME_EXCEEDED: 400 err = EHOSTUNREACH; 401 break; 402 default: 403 goto out; 404 } 405 406 switch (sk->sk_state) { 407 struct request_sock *req, **prev; 408 case TCP_LISTEN: 409 if (sock_owned_by_user(sk)) 410 goto out; 411 412 req = inet_csk_search_req(sk, &prev, th->dest, 413 iph->daddr, iph->saddr); 414 if (!req) 415 goto out; 416 417 /* ICMPs are not backlogged, hence we cannot get 418 an established socket here. 419 */ 420 BUG_TRAP(!req->sk); 421 422 if (seq != tcp_rsk(req)->snt_isn) { 423 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); 424 goto out; 425 } 426 427 /* 428 * Still in SYN_RECV, just remove it silently. 429 * There is no good way to pass the error to the newly 430 * created socket, and POSIX does not want network 431 * errors returned from accept(). 432 */ 433 inet_csk_reqsk_queue_drop(sk, req, prev); 434 goto out; 435 436 case TCP_SYN_SENT: 437 case TCP_SYN_RECV: /* Cannot happen. 438 It can f.e. if SYNs crossed. 439 */ 440 if (!sock_owned_by_user(sk)) { 441 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); 442 sk->sk_err = err; 443 444 sk->sk_error_report(sk); 445 446 tcp_done(sk); 447 } else { 448 sk->sk_err_soft = err; 449 } 450 goto out; 451 } 452 453 /* If we've already connected we will keep trying 454 * until we time out, or the user gives up. 455 * 456 * rfc1122 4.2.3.9 allows to consider as hard errors 457 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, 458 * but it is obsoleted by pmtu discovery). 459 * 460 * Note, that in modern internet, where routing is unreliable 461 * and in each dark corner broken firewalls sit, sending random 462 * errors ordered by their masters even this two messages finally lose 463 * their original sense (even Linux sends invalid PORT_UNREACHs) 464 * 465 * Now we are in compliance with RFCs. 466 * --ANK (980905) 467 */ 468 469 inet = inet_sk(sk); 470 if (!sock_owned_by_user(sk) && inet->recverr) { 471 sk->sk_err = err; 472 sk->sk_error_report(sk); 473 } else { /* Only an error on timeout */ 474 sk->sk_err_soft = err; 475 } 476 477 out: 478 bh_unlock_sock(sk); 479 sock_put(sk); 480 } 481 482 /* This routine computes an IPv4 TCP checksum. */ 483 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb) 484 { 485 struct inet_sock *inet = inet_sk(sk); 486 struct tcphdr *th = skb->h.th; 487 488 if (skb->ip_summed == CHECKSUM_HW) { 489 th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0); 490 skb->csum = offsetof(struct tcphdr, check); 491 } else { 492 th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr, 493 csum_partial((char *)th, 494 th->doff << 2, 495 skb->csum)); 496 } 497 } 498 499 int tcp_v4_gso_send_check(struct sk_buff *skb) 500 { 501 struct iphdr *iph; 502 struct tcphdr *th; 503 504 if (!pskb_may_pull(skb, sizeof(*th))) 505 return -EINVAL; 506 507 iph = skb->nh.iph; 508 th = skb->h.th; 509 510 th->check = 0; 511 th->check = ~tcp_v4_check(th, skb->len, iph->saddr, iph->daddr, 0); 512 skb->csum = offsetof(struct tcphdr, check); 513 skb->ip_summed = CHECKSUM_HW; 514 return 0; 515 } 516 517 /* 518 * This routine will send an RST to the other tcp. 519 * 520 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) 521 * for reset. 522 * Answer: if a packet caused RST, it is not for a socket 523 * existing in our system, if it is matched to a socket, 524 * it is just duplicate segment or bug in other side's TCP. 525 * So that we build reply only basing on parameters 526 * arrived with segment. 527 * Exception: precedence violation. We do not implement it in any case. 528 */ 529 530 static void tcp_v4_send_reset(struct sk_buff *skb) 531 { 532 struct tcphdr *th = skb->h.th; 533 struct tcphdr rth; 534 struct ip_reply_arg arg; 535 536 /* Never send a reset in response to a reset. */ 537 if (th->rst) 538 return; 539 540 if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL) 541 return; 542 543 /* Swap the send and the receive. */ 544 memset(&rth, 0, sizeof(struct tcphdr)); 545 rth.dest = th->source; 546 rth.source = th->dest; 547 rth.doff = sizeof(struct tcphdr) / 4; 548 rth.rst = 1; 549 550 if (th->ack) { 551 rth.seq = th->ack_seq; 552 } else { 553 rth.ack = 1; 554 rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + 555 skb->len - (th->doff << 2)); 556 } 557 558 memset(&arg, 0, sizeof arg); 559 arg.iov[0].iov_base = (unsigned char *)&rth; 560 arg.iov[0].iov_len = sizeof rth; 561 arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, 562 skb->nh.iph->saddr, /*XXX*/ 563 sizeof(struct tcphdr), IPPROTO_TCP, 0); 564 arg.csumoffset = offsetof(struct tcphdr, check) / 2; 565 566 ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth); 567 568 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); 569 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); 570 } 571 572 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states 573 outside socket context is ugly, certainly. What can I do? 574 */ 575 576 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, 577 u32 win, u32 ts) 578 { 579 struct tcphdr *th = skb->h.th; 580 struct { 581 struct tcphdr th; 582 u32 tsopt[3]; 583 } rep; 584 struct ip_reply_arg arg; 585 586 memset(&rep.th, 0, sizeof(struct tcphdr)); 587 memset(&arg, 0, sizeof arg); 588 589 arg.iov[0].iov_base = (unsigned char *)&rep; 590 arg.iov[0].iov_len = sizeof(rep.th); 591 if (ts) { 592 rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 593 (TCPOPT_TIMESTAMP << 8) | 594 TCPOLEN_TIMESTAMP); 595 rep.tsopt[1] = htonl(tcp_time_stamp); 596 rep.tsopt[2] = htonl(ts); 597 arg.iov[0].iov_len = sizeof(rep); 598 } 599 600 /* Swap the send and the receive. */ 601 rep.th.dest = th->source; 602 rep.th.source = th->dest; 603 rep.th.doff = arg.iov[0].iov_len / 4; 604 rep.th.seq = htonl(seq); 605 rep.th.ack_seq = htonl(ack); 606 rep.th.ack = 1; 607 rep.th.window = htons(win); 608 609 arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, 610 skb->nh.iph->saddr, /*XXX*/ 611 arg.iov[0].iov_len, IPPROTO_TCP, 0); 612 arg.csumoffset = offsetof(struct tcphdr, check) / 2; 613 614 ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len); 615 616 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); 617 } 618 619 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) 620 { 621 struct inet_timewait_sock *tw = inet_twsk(sk); 622 const struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 623 624 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 625 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent); 626 627 inet_twsk_put(tw); 628 } 629 630 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req) 631 { 632 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, 633 req->ts_recent); 634 } 635 636 /* 637 * Send a SYN-ACK after having received an ACK. 638 * This still operates on a request_sock only, not on a big 639 * socket. 640 */ 641 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req, 642 struct dst_entry *dst) 643 { 644 const struct inet_request_sock *ireq = inet_rsk(req); 645 int err = -1; 646 struct sk_buff * skb; 647 648 /* First, grab a route. */ 649 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) 650 goto out; 651 652 skb = tcp_make_synack(sk, dst, req); 653 654 if (skb) { 655 struct tcphdr *th = skb->h.th; 656 657 th->check = tcp_v4_check(th, skb->len, 658 ireq->loc_addr, 659 ireq->rmt_addr, 660 csum_partial((char *)th, skb->len, 661 skb->csum)); 662 663 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, 664 ireq->rmt_addr, 665 ireq->opt); 666 if (err == NET_XMIT_CN) 667 err = 0; 668 } 669 670 out: 671 dst_release(dst); 672 return err; 673 } 674 675 /* 676 * IPv4 request_sock destructor. 677 */ 678 static void tcp_v4_reqsk_destructor(struct request_sock *req) 679 { 680 kfree(inet_rsk(req)->opt); 681 } 682 683 #ifdef CONFIG_SYN_COOKIES 684 static void syn_flood_warning(struct sk_buff *skb) 685 { 686 static unsigned long warntime; 687 688 if (time_after(jiffies, (warntime + HZ * 60))) { 689 warntime = jiffies; 690 printk(KERN_INFO 691 "possible SYN flooding on port %d. Sending cookies.\n", 692 ntohs(skb->h.th->dest)); 693 } 694 } 695 #endif 696 697 /* 698 * Save and compile IPv4 options into the request_sock if needed. 699 */ 700 static struct ip_options *tcp_v4_save_options(struct sock *sk, 701 struct sk_buff *skb) 702 { 703 struct ip_options *opt = &(IPCB(skb)->opt); 704 struct ip_options *dopt = NULL; 705 706 if (opt && opt->optlen) { 707 int opt_size = optlength(opt); 708 dopt = kmalloc(opt_size, GFP_ATOMIC); 709 if (dopt) { 710 if (ip_options_echo(dopt, skb)) { 711 kfree(dopt); 712 dopt = NULL; 713 } 714 } 715 } 716 return dopt; 717 } 718 719 struct request_sock_ops tcp_request_sock_ops = { 720 .family = PF_INET, 721 .obj_size = sizeof(struct tcp_request_sock), 722 .rtx_syn_ack = tcp_v4_send_synack, 723 .send_ack = tcp_v4_reqsk_send_ack, 724 .destructor = tcp_v4_reqsk_destructor, 725 .send_reset = tcp_v4_send_reset, 726 }; 727 728 static struct timewait_sock_ops tcp_timewait_sock_ops = { 729 .twsk_obj_size = sizeof(struct tcp_timewait_sock), 730 .twsk_unique = tcp_twsk_unique, 731 }; 732 733 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) 734 { 735 struct inet_request_sock *ireq; 736 struct tcp_options_received tmp_opt; 737 struct request_sock *req; 738 __u32 saddr = skb->nh.iph->saddr; 739 __u32 daddr = skb->nh.iph->daddr; 740 __u32 isn = TCP_SKB_CB(skb)->when; 741 struct dst_entry *dst = NULL; 742 #ifdef CONFIG_SYN_COOKIES 743 int want_cookie = 0; 744 #else 745 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ 746 #endif 747 748 /* Never answer to SYNs send to broadcast or multicast */ 749 if (((struct rtable *)skb->dst)->rt_flags & 750 (RTCF_BROADCAST | RTCF_MULTICAST)) 751 goto drop; 752 753 /* TW buckets are converted to open requests without 754 * limitations, they conserve resources and peer is 755 * evidently real one. 756 */ 757 if (inet_csk_reqsk_queue_is_full(sk) && !isn) { 758 #ifdef CONFIG_SYN_COOKIES 759 if (sysctl_tcp_syncookies) { 760 want_cookie = 1; 761 } else 762 #endif 763 goto drop; 764 } 765 766 /* Accept backlog is full. If we have already queued enough 767 * of warm entries in syn queue, drop request. It is better than 768 * clogging syn queue with openreqs with exponentially increasing 769 * timeout. 770 */ 771 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) 772 goto drop; 773 774 req = reqsk_alloc(&tcp_request_sock_ops); 775 if (!req) 776 goto drop; 777 778 tcp_clear_options(&tmp_opt); 779 tmp_opt.mss_clamp = 536; 780 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; 781 782 tcp_parse_options(skb, &tmp_opt, 0); 783 784 if (want_cookie) { 785 tcp_clear_options(&tmp_opt); 786 tmp_opt.saw_tstamp = 0; 787 } 788 789 if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { 790 /* Some OSes (unknown ones, but I see them on web server, which 791 * contains information interesting only for windows' 792 * users) do not send their stamp in SYN. It is easy case. 793 * We simply do not advertise TS support. 794 */ 795 tmp_opt.saw_tstamp = 0; 796 tmp_opt.tstamp_ok = 0; 797 } 798 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; 799 800 tcp_openreq_init(req, &tmp_opt, skb); 801 802 ireq = inet_rsk(req); 803 ireq->loc_addr = daddr; 804 ireq->rmt_addr = saddr; 805 ireq->opt = tcp_v4_save_options(sk, skb); 806 if (!want_cookie) 807 TCP_ECN_create_request(req, skb->h.th); 808 809 if (want_cookie) { 810 #ifdef CONFIG_SYN_COOKIES 811 syn_flood_warning(skb); 812 #endif 813 isn = cookie_v4_init_sequence(sk, skb, &req->mss); 814 } else if (!isn) { 815 struct inet_peer *peer = NULL; 816 817 /* VJ's idea. We save last timestamp seen 818 * from the destination in peer table, when entering 819 * state TIME-WAIT, and check against it before 820 * accepting new connection request. 821 * 822 * If "isn" is not zero, this request hit alive 823 * timewait bucket, so that all the necessary checks 824 * are made in the function processing timewait state. 825 */ 826 if (tmp_opt.saw_tstamp && 827 tcp_death_row.sysctl_tw_recycle && 828 (dst = inet_csk_route_req(sk, req)) != NULL && 829 (peer = rt_get_peer((struct rtable *)dst)) != NULL && 830 peer->v4daddr == saddr) { 831 if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL && 832 (s32)(peer->tcp_ts - req->ts_recent) > 833 TCP_PAWS_WINDOW) { 834 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); 835 dst_release(dst); 836 goto drop_and_free; 837 } 838 } 839 /* Kill the following clause, if you dislike this way. */ 840 else if (!sysctl_tcp_syncookies && 841 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < 842 (sysctl_max_syn_backlog >> 2)) && 843 (!peer || !peer->tcp_ts_stamp) && 844 (!dst || !dst_metric(dst, RTAX_RTT))) { 845 /* Without syncookies last quarter of 846 * backlog is filled with destinations, 847 * proven to be alive. 848 * It means that we continue to communicate 849 * to destinations, already remembered 850 * to the moment of synflood. 851 */ 852 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open " 853 "request from %u.%u.%u.%u/%u\n", 854 NIPQUAD(saddr), 855 ntohs(skb->h.th->source)); 856 dst_release(dst); 857 goto drop_and_free; 858 } 859 860 isn = tcp_v4_init_sequence(sk, skb); 861 } 862 tcp_rsk(req)->snt_isn = isn; 863 864 if (tcp_v4_send_synack(sk, req, dst)) 865 goto drop_and_free; 866 867 if (want_cookie) { 868 reqsk_free(req); 869 } else { 870 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); 871 } 872 return 0; 873 874 drop_and_free: 875 reqsk_free(req); 876 drop: 877 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); 878 return 0; 879 } 880 881 882 /* 883 * The three way handshake has completed - we got a valid synack - 884 * now create the new socket. 885 */ 886 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 887 struct request_sock *req, 888 struct dst_entry *dst) 889 { 890 struct inet_request_sock *ireq; 891 struct inet_sock *newinet; 892 struct tcp_sock *newtp; 893 struct sock *newsk; 894 895 if (sk_acceptq_is_full(sk)) 896 goto exit_overflow; 897 898 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) 899 goto exit; 900 901 newsk = tcp_create_openreq_child(sk, req, skb); 902 if (!newsk) 903 goto exit; 904 905 newsk->sk_gso_type = SKB_GSO_TCPV4; 906 sk_setup_caps(newsk, dst); 907 908 newtp = tcp_sk(newsk); 909 newinet = inet_sk(newsk); 910 ireq = inet_rsk(req); 911 newinet->daddr = ireq->rmt_addr; 912 newinet->rcv_saddr = ireq->loc_addr; 913 newinet->saddr = ireq->loc_addr; 914 newinet->opt = ireq->opt; 915 ireq->opt = NULL; 916 newinet->mc_index = inet_iif(skb); 917 newinet->mc_ttl = skb->nh.iph->ttl; 918 inet_csk(newsk)->icsk_ext_hdr_len = 0; 919 if (newinet->opt) 920 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen; 921 newinet->id = newtp->write_seq ^ jiffies; 922 923 tcp_mtup_init(newsk); 924 tcp_sync_mss(newsk, dst_mtu(dst)); 925 newtp->advmss = dst_metric(dst, RTAX_ADVMSS); 926 tcp_initialize_rcv_mss(newsk); 927 928 __inet_hash(&tcp_hashinfo, newsk, 0); 929 __inet_inherit_port(&tcp_hashinfo, sk, newsk); 930 931 return newsk; 932 933 exit_overflow: 934 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); 935 exit: 936 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); 937 dst_release(dst); 938 return NULL; 939 } 940 941 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) 942 { 943 struct tcphdr *th = skb->h.th; 944 struct iphdr *iph = skb->nh.iph; 945 struct sock *nsk; 946 struct request_sock **prev; 947 /* Find possible connection requests. */ 948 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, 949 iph->saddr, iph->daddr); 950 if (req) 951 return tcp_check_req(sk, skb, req, prev); 952 953 nsk = __inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr, 954 th->source, skb->nh.iph->daddr, 955 ntohs(th->dest), inet_iif(skb)); 956 957 if (nsk) { 958 if (nsk->sk_state != TCP_TIME_WAIT) { 959 bh_lock_sock(nsk); 960 return nsk; 961 } 962 inet_twsk_put((struct inet_timewait_sock *)nsk); 963 return NULL; 964 } 965 966 #ifdef CONFIG_SYN_COOKIES 967 if (!th->rst && !th->syn && th->ack) 968 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); 969 #endif 970 return sk; 971 } 972 973 static int tcp_v4_checksum_init(struct sk_buff *skb) 974 { 975 if (skb->ip_summed == CHECKSUM_HW) { 976 if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, 977 skb->nh.iph->daddr, skb->csum)) { 978 skb->ip_summed = CHECKSUM_UNNECESSARY; 979 return 0; 980 } 981 } 982 983 skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr, 984 skb->len, IPPROTO_TCP, 0); 985 986 if (skb->len <= 76) { 987 return __skb_checksum_complete(skb); 988 } 989 return 0; 990 } 991 992 993 /* The socket must have it's spinlock held when we get 994 * here. 995 * 996 * We have a potential double-lock case here, so even when 997 * doing backlog processing we use the BH locking scheme. 998 * This is because we cannot sleep with the original spinlock 999 * held. 1000 */ 1001 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) 1002 { 1003 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1004 TCP_CHECK_TIMER(sk); 1005 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) 1006 goto reset; 1007 TCP_CHECK_TIMER(sk); 1008 return 0; 1009 } 1010 1011 if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb)) 1012 goto csum_err; 1013 1014 if (sk->sk_state == TCP_LISTEN) { 1015 struct sock *nsk = tcp_v4_hnd_req(sk, skb); 1016 if (!nsk) 1017 goto discard; 1018 1019 if (nsk != sk) { 1020 if (tcp_child_process(sk, nsk, skb)) 1021 goto reset; 1022 return 0; 1023 } 1024 } 1025 1026 TCP_CHECK_TIMER(sk); 1027 if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) 1028 goto reset; 1029 TCP_CHECK_TIMER(sk); 1030 return 0; 1031 1032 reset: 1033 tcp_v4_send_reset(skb); 1034 discard: 1035 kfree_skb(skb); 1036 /* Be careful here. If this function gets more complicated and 1037 * gcc suffers from register pressure on the x86, sk (in %ebx) 1038 * might be destroyed here. This current version compiles correctly, 1039 * but you have been warned. 1040 */ 1041 return 0; 1042 1043 csum_err: 1044 TCP_INC_STATS_BH(TCP_MIB_INERRS); 1045 goto discard; 1046 } 1047 1048 /* 1049 * From tcp_input.c 1050 */ 1051 1052 int tcp_v4_rcv(struct sk_buff *skb) 1053 { 1054 struct tcphdr *th; 1055 struct sock *sk; 1056 int ret; 1057 1058 if (skb->pkt_type != PACKET_HOST) 1059 goto discard_it; 1060 1061 /* Count it even if it's bad */ 1062 TCP_INC_STATS_BH(TCP_MIB_INSEGS); 1063 1064 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1065 goto discard_it; 1066 1067 th = skb->h.th; 1068 1069 if (th->doff < sizeof(struct tcphdr) / 4) 1070 goto bad_packet; 1071 if (!pskb_may_pull(skb, th->doff * 4)) 1072 goto discard_it; 1073 1074 /* An explanation is required here, I think. 1075 * Packet length and doff are validated by header prediction, 1076 * provided case of th->doff==0 is eliminated. 1077 * So, we defer the checks. */ 1078 if ((skb->ip_summed != CHECKSUM_UNNECESSARY && 1079 tcp_v4_checksum_init(skb))) 1080 goto bad_packet; 1081 1082 th = skb->h.th; 1083 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1084 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1085 skb->len - th->doff * 4); 1086 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1087 TCP_SKB_CB(skb)->when = 0; 1088 TCP_SKB_CB(skb)->flags = skb->nh.iph->tos; 1089 TCP_SKB_CB(skb)->sacked = 0; 1090 1091 sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source, 1092 skb->nh.iph->daddr, ntohs(th->dest), 1093 inet_iif(skb)); 1094 1095 if (!sk) 1096 goto no_tcp_socket; 1097 1098 process: 1099 if (sk->sk_state == TCP_TIME_WAIT) 1100 goto do_time_wait; 1101 1102 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1103 goto discard_and_relse; 1104 nf_reset(skb); 1105 1106 if (sk_filter(sk, skb, 0)) 1107 goto discard_and_relse; 1108 1109 skb->dev = NULL; 1110 1111 bh_lock_sock_nested(sk); 1112 ret = 0; 1113 if (!sock_owned_by_user(sk)) { 1114 #ifdef CONFIG_NET_DMA 1115 struct tcp_sock *tp = tcp_sk(sk); 1116 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) 1117 tp->ucopy.dma_chan = get_softnet_dma(); 1118 if (tp->ucopy.dma_chan) 1119 ret = tcp_v4_do_rcv(sk, skb); 1120 else 1121 #endif 1122 { 1123 if (!tcp_prequeue(sk, skb)) 1124 ret = tcp_v4_do_rcv(sk, skb); 1125 } 1126 } else 1127 sk_add_backlog(sk, skb); 1128 bh_unlock_sock(sk); 1129 1130 sock_put(sk); 1131 1132 return ret; 1133 1134 no_tcp_socket: 1135 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1136 goto discard_it; 1137 1138 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { 1139 bad_packet: 1140 TCP_INC_STATS_BH(TCP_MIB_INERRS); 1141 } else { 1142 tcp_v4_send_reset(skb); 1143 } 1144 1145 discard_it: 1146 /* Discard frame. */ 1147 kfree_skb(skb); 1148 return 0; 1149 1150 discard_and_relse: 1151 sock_put(sk); 1152 goto discard_it; 1153 1154 do_time_wait: 1155 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1156 inet_twsk_put((struct inet_timewait_sock *) sk); 1157 goto discard_it; 1158 } 1159 1160 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { 1161 TCP_INC_STATS_BH(TCP_MIB_INERRS); 1162 inet_twsk_put((struct inet_timewait_sock *) sk); 1163 goto discard_it; 1164 } 1165 switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk, 1166 skb, th)) { 1167 case TCP_TW_SYN: { 1168 struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo, 1169 skb->nh.iph->daddr, 1170 ntohs(th->dest), 1171 inet_iif(skb)); 1172 if (sk2) { 1173 inet_twsk_deschedule((struct inet_timewait_sock *)sk, 1174 &tcp_death_row); 1175 inet_twsk_put((struct inet_timewait_sock *)sk); 1176 sk = sk2; 1177 goto process; 1178 } 1179 /* Fall through to ACK */ 1180 } 1181 case TCP_TW_ACK: 1182 tcp_v4_timewait_ack(sk, skb); 1183 break; 1184 case TCP_TW_RST: 1185 goto no_tcp_socket; 1186 case TCP_TW_SUCCESS:; 1187 } 1188 goto discard_it; 1189 } 1190 1191 /* VJ's idea. Save last timestamp seen from this destination 1192 * and hold it at least for normal timewait interval to use for duplicate 1193 * segment detection in subsequent connections, before they enter synchronized 1194 * state. 1195 */ 1196 1197 int tcp_v4_remember_stamp(struct sock *sk) 1198 { 1199 struct inet_sock *inet = inet_sk(sk); 1200 struct tcp_sock *tp = tcp_sk(sk); 1201 struct rtable *rt = (struct rtable *)__sk_dst_get(sk); 1202 struct inet_peer *peer = NULL; 1203 int release_it = 0; 1204 1205 if (!rt || rt->rt_dst != inet->daddr) { 1206 peer = inet_getpeer(inet->daddr, 1); 1207 release_it = 1; 1208 } else { 1209 if (!rt->peer) 1210 rt_bind_peer(rt, 1); 1211 peer = rt->peer; 1212 } 1213 1214 if (peer) { 1215 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || 1216 (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && 1217 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { 1218 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; 1219 peer->tcp_ts = tp->rx_opt.ts_recent; 1220 } 1221 if (release_it) 1222 inet_putpeer(peer); 1223 return 1; 1224 } 1225 1226 return 0; 1227 } 1228 1229 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw) 1230 { 1231 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1); 1232 1233 if (peer) { 1234 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); 1235 1236 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 || 1237 (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && 1238 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) { 1239 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp; 1240 peer->tcp_ts = tcptw->tw_ts_recent; 1241 } 1242 inet_putpeer(peer); 1243 return 1; 1244 } 1245 1246 return 0; 1247 } 1248 1249 struct inet_connection_sock_af_ops ipv4_specific = { 1250 .queue_xmit = ip_queue_xmit, 1251 .send_check = tcp_v4_send_check, 1252 .rebuild_header = inet_sk_rebuild_header, 1253 .conn_request = tcp_v4_conn_request, 1254 .syn_recv_sock = tcp_v4_syn_recv_sock, 1255 .remember_stamp = tcp_v4_remember_stamp, 1256 .net_header_len = sizeof(struct iphdr), 1257 .setsockopt = ip_setsockopt, 1258 .getsockopt = ip_getsockopt, 1259 .addr2sockaddr = inet_csk_addr2sockaddr, 1260 .sockaddr_len = sizeof(struct sockaddr_in), 1261 #ifdef CONFIG_COMPAT 1262 .compat_setsockopt = compat_ip_setsockopt, 1263 .compat_getsockopt = compat_ip_getsockopt, 1264 #endif 1265 }; 1266 1267 /* NOTE: A lot of things set to zero explicitly by call to 1268 * sk_alloc() so need not be done here. 1269 */ 1270 static int tcp_v4_init_sock(struct sock *sk) 1271 { 1272 struct inet_connection_sock *icsk = inet_csk(sk); 1273 struct tcp_sock *tp = tcp_sk(sk); 1274 1275 skb_queue_head_init(&tp->out_of_order_queue); 1276 tcp_init_xmit_timers(sk); 1277 tcp_prequeue_init(tp); 1278 1279 icsk->icsk_rto = TCP_TIMEOUT_INIT; 1280 tp->mdev = TCP_TIMEOUT_INIT; 1281 1282 /* So many TCP implementations out there (incorrectly) count the 1283 * initial SYN frame in their delayed-ACK and congestion control 1284 * algorithms that we must have the following bandaid to talk 1285 * efficiently to them. -DaveM 1286 */ 1287 tp->snd_cwnd = 2; 1288 1289 /* See draft-stevens-tcpca-spec-01 for discussion of the 1290 * initialization of these values. 1291 */ 1292 tp->snd_ssthresh = 0x7fffffff; /* Infinity */ 1293 tp->snd_cwnd_clamp = ~0; 1294 tp->mss_cache = 536; 1295 1296 tp->reordering = sysctl_tcp_reordering; 1297 icsk->icsk_ca_ops = &tcp_init_congestion_ops; 1298 1299 sk->sk_state = TCP_CLOSE; 1300 1301 sk->sk_write_space = sk_stream_write_space; 1302 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1303 1304 icsk->icsk_af_ops = &ipv4_specific; 1305 icsk->icsk_sync_mss = tcp_sync_mss; 1306 1307 sk->sk_sndbuf = sysctl_tcp_wmem[1]; 1308 sk->sk_rcvbuf = sysctl_tcp_rmem[1]; 1309 1310 atomic_inc(&tcp_sockets_allocated); 1311 1312 return 0; 1313 } 1314 1315 int tcp_v4_destroy_sock(struct sock *sk) 1316 { 1317 struct tcp_sock *tp = tcp_sk(sk); 1318 1319 tcp_clear_xmit_timers(sk); 1320 1321 tcp_cleanup_congestion_control(sk); 1322 1323 /* Cleanup up the write buffer. */ 1324 sk_stream_writequeue_purge(sk); 1325 1326 /* Cleans up our, hopefully empty, out_of_order_queue. */ 1327 __skb_queue_purge(&tp->out_of_order_queue); 1328 1329 #ifdef CONFIG_NET_DMA 1330 /* Cleans up our sk_async_wait_queue */ 1331 __skb_queue_purge(&sk->sk_async_wait_queue); 1332 #endif 1333 1334 /* Clean prequeue, it must be empty really */ 1335 __skb_queue_purge(&tp->ucopy.prequeue); 1336 1337 /* Clean up a referenced TCP bind bucket. */ 1338 if (inet_csk(sk)->icsk_bind_hash) 1339 inet_put_port(&tcp_hashinfo, sk); 1340 1341 /* 1342 * If sendmsg cached page exists, toss it. 1343 */ 1344 if (sk->sk_sndmsg_page) { 1345 __free_page(sk->sk_sndmsg_page); 1346 sk->sk_sndmsg_page = NULL; 1347 } 1348 1349 atomic_dec(&tcp_sockets_allocated); 1350 1351 return 0; 1352 } 1353 1354 EXPORT_SYMBOL(tcp_v4_destroy_sock); 1355 1356 #ifdef CONFIG_PROC_FS 1357 /* Proc filesystem TCP sock list dumping. */ 1358 1359 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head) 1360 { 1361 return hlist_empty(head) ? NULL : 1362 list_entry(head->first, struct inet_timewait_sock, tw_node); 1363 } 1364 1365 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) 1366 { 1367 return tw->tw_node.next ? 1368 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; 1369 } 1370 1371 static void *listening_get_next(struct seq_file *seq, void *cur) 1372 { 1373 struct inet_connection_sock *icsk; 1374 struct hlist_node *node; 1375 struct sock *sk = cur; 1376 struct tcp_iter_state* st = seq->private; 1377 1378 if (!sk) { 1379 st->bucket = 0; 1380 sk = sk_head(&tcp_hashinfo.listening_hash[0]); 1381 goto get_sk; 1382 } 1383 1384 ++st->num; 1385 1386 if (st->state == TCP_SEQ_STATE_OPENREQ) { 1387 struct request_sock *req = cur; 1388 1389 icsk = inet_csk(st->syn_wait_sk); 1390 req = req->dl_next; 1391 while (1) { 1392 while (req) { 1393 if (req->rsk_ops->family == st->family) { 1394 cur = req; 1395 goto out; 1396 } 1397 req = req->dl_next; 1398 } 1399 if (++st->sbucket >= TCP_SYNQ_HSIZE) 1400 break; 1401 get_req: 1402 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; 1403 } 1404 sk = sk_next(st->syn_wait_sk); 1405 st->state = TCP_SEQ_STATE_LISTENING; 1406 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1407 } else { 1408 icsk = inet_csk(sk); 1409 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1410 if (reqsk_queue_len(&icsk->icsk_accept_queue)) 1411 goto start_req; 1412 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1413 sk = sk_next(sk); 1414 } 1415 get_sk: 1416 sk_for_each_from(sk, node) { 1417 if (sk->sk_family == st->family) { 1418 cur = sk; 1419 goto out; 1420 } 1421 icsk = inet_csk(sk); 1422 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1423 if (reqsk_queue_len(&icsk->icsk_accept_queue)) { 1424 start_req: 1425 st->uid = sock_i_uid(sk); 1426 st->syn_wait_sk = sk; 1427 st->state = TCP_SEQ_STATE_OPENREQ; 1428 st->sbucket = 0; 1429 goto get_req; 1430 } 1431 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1432 } 1433 if (++st->bucket < INET_LHTABLE_SIZE) { 1434 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]); 1435 goto get_sk; 1436 } 1437 cur = NULL; 1438 out: 1439 return cur; 1440 } 1441 1442 static void *listening_get_idx(struct seq_file *seq, loff_t *pos) 1443 { 1444 void *rc = listening_get_next(seq, NULL); 1445 1446 while (rc && *pos) { 1447 rc = listening_get_next(seq, rc); 1448 --*pos; 1449 } 1450 return rc; 1451 } 1452 1453 static void *established_get_first(struct seq_file *seq) 1454 { 1455 struct tcp_iter_state* st = seq->private; 1456 void *rc = NULL; 1457 1458 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) { 1459 struct sock *sk; 1460 struct hlist_node *node; 1461 struct inet_timewait_sock *tw; 1462 1463 /* We can reschedule _before_ having picked the target: */ 1464 cond_resched_softirq(); 1465 1466 read_lock(&tcp_hashinfo.ehash[st->bucket].lock); 1467 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { 1468 if (sk->sk_family != st->family) { 1469 continue; 1470 } 1471 rc = sk; 1472 goto out; 1473 } 1474 st->state = TCP_SEQ_STATE_TIME_WAIT; 1475 inet_twsk_for_each(tw, node, 1476 &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) { 1477 if (tw->tw_family != st->family) { 1478 continue; 1479 } 1480 rc = tw; 1481 goto out; 1482 } 1483 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); 1484 st->state = TCP_SEQ_STATE_ESTABLISHED; 1485 } 1486 out: 1487 return rc; 1488 } 1489 1490 static void *established_get_next(struct seq_file *seq, void *cur) 1491 { 1492 struct sock *sk = cur; 1493 struct inet_timewait_sock *tw; 1494 struct hlist_node *node; 1495 struct tcp_iter_state* st = seq->private; 1496 1497 ++st->num; 1498 1499 if (st->state == TCP_SEQ_STATE_TIME_WAIT) { 1500 tw = cur; 1501 tw = tw_next(tw); 1502 get_tw: 1503 while (tw && tw->tw_family != st->family) { 1504 tw = tw_next(tw); 1505 } 1506 if (tw) { 1507 cur = tw; 1508 goto out; 1509 } 1510 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); 1511 st->state = TCP_SEQ_STATE_ESTABLISHED; 1512 1513 /* We can reschedule between buckets: */ 1514 cond_resched_softirq(); 1515 1516 if (++st->bucket < tcp_hashinfo.ehash_size) { 1517 read_lock(&tcp_hashinfo.ehash[st->bucket].lock); 1518 sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain); 1519 } else { 1520 cur = NULL; 1521 goto out; 1522 } 1523 } else 1524 sk = sk_next(sk); 1525 1526 sk_for_each_from(sk, node) { 1527 if (sk->sk_family == st->family) 1528 goto found; 1529 } 1530 1531 st->state = TCP_SEQ_STATE_TIME_WAIT; 1532 tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain); 1533 goto get_tw; 1534 found: 1535 cur = sk; 1536 out: 1537 return cur; 1538 } 1539 1540 static void *established_get_idx(struct seq_file *seq, loff_t pos) 1541 { 1542 void *rc = established_get_first(seq); 1543 1544 while (rc && pos) { 1545 rc = established_get_next(seq, rc); 1546 --pos; 1547 } 1548 return rc; 1549 } 1550 1551 static void *tcp_get_idx(struct seq_file *seq, loff_t pos) 1552 { 1553 void *rc; 1554 struct tcp_iter_state* st = seq->private; 1555 1556 inet_listen_lock(&tcp_hashinfo); 1557 st->state = TCP_SEQ_STATE_LISTENING; 1558 rc = listening_get_idx(seq, &pos); 1559 1560 if (!rc) { 1561 inet_listen_unlock(&tcp_hashinfo); 1562 local_bh_disable(); 1563 st->state = TCP_SEQ_STATE_ESTABLISHED; 1564 rc = established_get_idx(seq, pos); 1565 } 1566 1567 return rc; 1568 } 1569 1570 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) 1571 { 1572 struct tcp_iter_state* st = seq->private; 1573 st->state = TCP_SEQ_STATE_LISTENING; 1574 st->num = 0; 1575 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1576 } 1577 1578 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1579 { 1580 void *rc = NULL; 1581 struct tcp_iter_state* st; 1582 1583 if (v == SEQ_START_TOKEN) { 1584 rc = tcp_get_idx(seq, 0); 1585 goto out; 1586 } 1587 st = seq->private; 1588 1589 switch (st->state) { 1590 case TCP_SEQ_STATE_OPENREQ: 1591 case TCP_SEQ_STATE_LISTENING: 1592 rc = listening_get_next(seq, v); 1593 if (!rc) { 1594 inet_listen_unlock(&tcp_hashinfo); 1595 local_bh_disable(); 1596 st->state = TCP_SEQ_STATE_ESTABLISHED; 1597 rc = established_get_first(seq); 1598 } 1599 break; 1600 case TCP_SEQ_STATE_ESTABLISHED: 1601 case TCP_SEQ_STATE_TIME_WAIT: 1602 rc = established_get_next(seq, v); 1603 break; 1604 } 1605 out: 1606 ++*pos; 1607 return rc; 1608 } 1609 1610 static void tcp_seq_stop(struct seq_file *seq, void *v) 1611 { 1612 struct tcp_iter_state* st = seq->private; 1613 1614 switch (st->state) { 1615 case TCP_SEQ_STATE_OPENREQ: 1616 if (v) { 1617 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); 1618 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1619 } 1620 case TCP_SEQ_STATE_LISTENING: 1621 if (v != SEQ_START_TOKEN) 1622 inet_listen_unlock(&tcp_hashinfo); 1623 break; 1624 case TCP_SEQ_STATE_TIME_WAIT: 1625 case TCP_SEQ_STATE_ESTABLISHED: 1626 if (v) 1627 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); 1628 local_bh_enable(); 1629 break; 1630 } 1631 } 1632 1633 static int tcp_seq_open(struct inode *inode, struct file *file) 1634 { 1635 struct tcp_seq_afinfo *afinfo = PDE(inode)->data; 1636 struct seq_file *seq; 1637 struct tcp_iter_state *s; 1638 int rc; 1639 1640 if (unlikely(afinfo == NULL)) 1641 return -EINVAL; 1642 1643 s = kmalloc(sizeof(*s), GFP_KERNEL); 1644 if (!s) 1645 return -ENOMEM; 1646 memset(s, 0, sizeof(*s)); 1647 s->family = afinfo->family; 1648 s->seq_ops.start = tcp_seq_start; 1649 s->seq_ops.next = tcp_seq_next; 1650 s->seq_ops.show = afinfo->seq_show; 1651 s->seq_ops.stop = tcp_seq_stop; 1652 1653 rc = seq_open(file, &s->seq_ops); 1654 if (rc) 1655 goto out_kfree; 1656 seq = file->private_data; 1657 seq->private = s; 1658 out: 1659 return rc; 1660 out_kfree: 1661 kfree(s); 1662 goto out; 1663 } 1664 1665 int tcp_proc_register(struct tcp_seq_afinfo *afinfo) 1666 { 1667 int rc = 0; 1668 struct proc_dir_entry *p; 1669 1670 if (!afinfo) 1671 return -EINVAL; 1672 afinfo->seq_fops->owner = afinfo->owner; 1673 afinfo->seq_fops->open = tcp_seq_open; 1674 afinfo->seq_fops->read = seq_read; 1675 afinfo->seq_fops->llseek = seq_lseek; 1676 afinfo->seq_fops->release = seq_release_private; 1677 1678 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); 1679 if (p) 1680 p->data = afinfo; 1681 else 1682 rc = -ENOMEM; 1683 return rc; 1684 } 1685 1686 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo) 1687 { 1688 if (!afinfo) 1689 return; 1690 proc_net_remove(afinfo->name); 1691 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); 1692 } 1693 1694 static void get_openreq4(struct sock *sk, struct request_sock *req, 1695 char *tmpbuf, int i, int uid) 1696 { 1697 const struct inet_request_sock *ireq = inet_rsk(req); 1698 int ttd = req->expires - jiffies; 1699 1700 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" 1701 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p", 1702 i, 1703 ireq->loc_addr, 1704 ntohs(inet_sk(sk)->sport), 1705 ireq->rmt_addr, 1706 ntohs(ireq->rmt_port), 1707 TCP_SYN_RECV, 1708 0, 0, /* could print option size, but that is af dependent. */ 1709 1, /* timers active (only the expire timer) */ 1710 jiffies_to_clock_t(ttd), 1711 req->retrans, 1712 uid, 1713 0, /* non standard timer */ 1714 0, /* open_requests have no inode */ 1715 atomic_read(&sk->sk_refcnt), 1716 req); 1717 } 1718 1719 static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i) 1720 { 1721 int timer_active; 1722 unsigned long timer_expires; 1723 struct tcp_sock *tp = tcp_sk(sp); 1724 const struct inet_connection_sock *icsk = inet_csk(sp); 1725 struct inet_sock *inet = inet_sk(sp); 1726 unsigned int dest = inet->daddr; 1727 unsigned int src = inet->rcv_saddr; 1728 __u16 destp = ntohs(inet->dport); 1729 __u16 srcp = ntohs(inet->sport); 1730 1731 if (icsk->icsk_pending == ICSK_TIME_RETRANS) { 1732 timer_active = 1; 1733 timer_expires = icsk->icsk_timeout; 1734 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 1735 timer_active = 4; 1736 timer_expires = icsk->icsk_timeout; 1737 } else if (timer_pending(&sp->sk_timer)) { 1738 timer_active = 2; 1739 timer_expires = sp->sk_timer.expires; 1740 } else { 1741 timer_active = 0; 1742 timer_expires = jiffies; 1743 } 1744 1745 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " 1746 "%08X %5d %8d %lu %d %p %u %u %u %u %d", 1747 i, src, srcp, dest, destp, sp->sk_state, 1748 tp->write_seq - tp->snd_una, 1749 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq), 1750 timer_active, 1751 jiffies_to_clock_t(timer_expires - jiffies), 1752 icsk->icsk_retransmits, 1753 sock_i_uid(sp), 1754 icsk->icsk_probes_out, 1755 sock_i_ino(sp), 1756 atomic_read(&sp->sk_refcnt), sp, 1757 icsk->icsk_rto, 1758 icsk->icsk_ack.ato, 1759 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, 1760 tp->snd_cwnd, 1761 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh); 1762 } 1763 1764 static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i) 1765 { 1766 unsigned int dest, src; 1767 __u16 destp, srcp; 1768 int ttd = tw->tw_ttd - jiffies; 1769 1770 if (ttd < 0) 1771 ttd = 0; 1772 1773 dest = tw->tw_daddr; 1774 src = tw->tw_rcv_saddr; 1775 destp = ntohs(tw->tw_dport); 1776 srcp = ntohs(tw->tw_sport); 1777 1778 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" 1779 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p", 1780 i, src, srcp, dest, destp, tw->tw_substate, 0, 0, 1781 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, 1782 atomic_read(&tw->tw_refcnt), tw); 1783 } 1784 1785 #define TMPSZ 150 1786 1787 static int tcp4_seq_show(struct seq_file *seq, void *v) 1788 { 1789 struct tcp_iter_state* st; 1790 char tmpbuf[TMPSZ + 1]; 1791 1792 if (v == SEQ_START_TOKEN) { 1793 seq_printf(seq, "%-*s\n", TMPSZ - 1, 1794 " sl local_address rem_address st tx_queue " 1795 "rx_queue tr tm->when retrnsmt uid timeout " 1796 "inode"); 1797 goto out; 1798 } 1799 st = seq->private; 1800 1801 switch (st->state) { 1802 case TCP_SEQ_STATE_LISTENING: 1803 case TCP_SEQ_STATE_ESTABLISHED: 1804 get_tcp4_sock(v, tmpbuf, st->num); 1805 break; 1806 case TCP_SEQ_STATE_OPENREQ: 1807 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid); 1808 break; 1809 case TCP_SEQ_STATE_TIME_WAIT: 1810 get_timewait4_sock(v, tmpbuf, st->num); 1811 break; 1812 } 1813 seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf); 1814 out: 1815 return 0; 1816 } 1817 1818 static struct file_operations tcp4_seq_fops; 1819 static struct tcp_seq_afinfo tcp4_seq_afinfo = { 1820 .owner = THIS_MODULE, 1821 .name = "tcp", 1822 .family = AF_INET, 1823 .seq_show = tcp4_seq_show, 1824 .seq_fops = &tcp4_seq_fops, 1825 }; 1826 1827 int __init tcp4_proc_init(void) 1828 { 1829 return tcp_proc_register(&tcp4_seq_afinfo); 1830 } 1831 1832 void tcp4_proc_exit(void) 1833 { 1834 tcp_proc_unregister(&tcp4_seq_afinfo); 1835 } 1836 #endif /* CONFIG_PROC_FS */ 1837 1838 struct proto tcp_prot = { 1839 .name = "TCP", 1840 .owner = THIS_MODULE, 1841 .close = tcp_close, 1842 .connect = tcp_v4_connect, 1843 .disconnect = tcp_disconnect, 1844 .accept = inet_csk_accept, 1845 .ioctl = tcp_ioctl, 1846 .init = tcp_v4_init_sock, 1847 .destroy = tcp_v4_destroy_sock, 1848 .shutdown = tcp_shutdown, 1849 .setsockopt = tcp_setsockopt, 1850 .getsockopt = tcp_getsockopt, 1851 .sendmsg = tcp_sendmsg, 1852 .recvmsg = tcp_recvmsg, 1853 .backlog_rcv = tcp_v4_do_rcv, 1854 .hash = tcp_v4_hash, 1855 .unhash = tcp_unhash, 1856 .get_port = tcp_v4_get_port, 1857 .enter_memory_pressure = tcp_enter_memory_pressure, 1858 .sockets_allocated = &tcp_sockets_allocated, 1859 .orphan_count = &tcp_orphan_count, 1860 .memory_allocated = &tcp_memory_allocated, 1861 .memory_pressure = &tcp_memory_pressure, 1862 .sysctl_mem = sysctl_tcp_mem, 1863 .sysctl_wmem = sysctl_tcp_wmem, 1864 .sysctl_rmem = sysctl_tcp_rmem, 1865 .max_header = MAX_TCP_HEADER, 1866 .obj_size = sizeof(struct tcp_sock), 1867 .twsk_prot = &tcp_timewait_sock_ops, 1868 .rsk_prot = &tcp_request_sock_ops, 1869 #ifdef CONFIG_COMPAT 1870 .compat_setsockopt = compat_tcp_setsockopt, 1871 .compat_getsockopt = compat_tcp_getsockopt, 1872 #endif 1873 }; 1874 1875 void __init tcp_v4_init(struct net_proto_family *ops) 1876 { 1877 if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW, IPPROTO_TCP) < 0) 1878 panic("Failed to create the TCP control socket.\n"); 1879 } 1880 1881 EXPORT_SYMBOL(ipv4_specific); 1882 EXPORT_SYMBOL(tcp_hashinfo); 1883 EXPORT_SYMBOL(tcp_prot); 1884 EXPORT_SYMBOL(tcp_unhash); 1885 EXPORT_SYMBOL(tcp_v4_conn_request); 1886 EXPORT_SYMBOL(tcp_v4_connect); 1887 EXPORT_SYMBOL(tcp_v4_do_rcv); 1888 EXPORT_SYMBOL(tcp_v4_remember_stamp); 1889 EXPORT_SYMBOL(tcp_v4_send_check); 1890 EXPORT_SYMBOL(tcp_v4_syn_recv_sock); 1891 1892 #ifdef CONFIG_PROC_FS 1893 EXPORT_SYMBOL(tcp_proc_register); 1894 EXPORT_SYMBOL(tcp_proc_unregister); 1895 #endif 1896 EXPORT_SYMBOL(sysctl_local_port_range); 1897 EXPORT_SYMBOL(sysctl_tcp_low_latency); 1898 1899