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 * IPv4 specific functions 9 * 10 * 11 * code split from: 12 * linux/ipv4/tcp.c 13 * linux/ipv4/tcp_input.c 14 * linux/ipv4/tcp_output.c 15 * 16 * See tcp.c for author information 17 * 18 * This program is free software; you can redistribute it and/or 19 * modify it under the terms of the GNU General Public License 20 * as published by the Free Software Foundation; either version 21 * 2 of the License, or (at your option) any later version. 22 */ 23 24 /* 25 * Changes: 26 * David S. Miller : New socket lookup architecture. 27 * This code is dedicated to John Dyson. 28 * David S. Miller : Change semantics of established hash, 29 * half is devoted to TIME_WAIT sockets 30 * and the rest go in the other half. 31 * Andi Kleen : Add support for syncookies and fixed 32 * some bugs: ip options weren't passed to 33 * the TCP layer, missed a check for an 34 * ACK bit. 35 * Andi Kleen : Implemented fast path mtu discovery. 36 * Fixed many serious bugs in the 37 * request_sock handling and moved 38 * most of it into the af independent code. 39 * Added tail drop and some other bugfixes. 40 * Added new listen semantics. 41 * Mike McLagan : Routing by source 42 * Juan Jose Ciarlante: ip_dynaddr bits 43 * Andi Kleen: various fixes. 44 * Vitaly E. Lavrov : Transparent proxy revived after year 45 * coma. 46 * Andi Kleen : Fix new listen. 47 * Andi Kleen : Fix accept error reporting. 48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 50 * a single port at the same time. 51 */ 52 53 #define pr_fmt(fmt) "TCP: " fmt 54 55 #include <linux/bottom_half.h> 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 #include <linux/slab.h> 65 66 #include <net/net_namespace.h> 67 #include <net/icmp.h> 68 #include <net/inet_hashtables.h> 69 #include <net/tcp.h> 70 #include <net/transp_v6.h> 71 #include <net/ipv6.h> 72 #include <net/inet_common.h> 73 #include <net/timewait_sock.h> 74 #include <net/xfrm.h> 75 #include <net/secure_seq.h> 76 #include <net/tcp_memcontrol.h> 77 #include <net/busy_poll.h> 78 79 #include <linux/inet.h> 80 #include <linux/ipv6.h> 81 #include <linux/stddef.h> 82 #include <linux/proc_fs.h> 83 #include <linux/seq_file.h> 84 85 #include <linux/crypto.h> 86 #include <linux/scatterlist.h> 87 88 int sysctl_tcp_tw_reuse __read_mostly; 89 int sysctl_tcp_low_latency __read_mostly; 90 EXPORT_SYMBOL(sysctl_tcp_low_latency); 91 92 #ifdef CONFIG_TCP_MD5SIG 93 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 94 __be32 daddr, __be32 saddr, const struct tcphdr *th); 95 #endif 96 97 struct inet_hashinfo tcp_hashinfo; 98 EXPORT_SYMBOL(tcp_hashinfo); 99 100 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb) 101 { 102 return secure_tcp_sequence_number(ip_hdr(skb)->daddr, 103 ip_hdr(skb)->saddr, 104 tcp_hdr(skb)->dest, 105 tcp_hdr(skb)->source); 106 } 107 108 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) 109 { 110 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); 111 struct tcp_sock *tp = tcp_sk(sk); 112 113 /* With PAWS, it is safe from the viewpoint 114 of data integrity. Even without PAWS it is safe provided sequence 115 spaces do not overlap i.e. at data rates <= 80Mbit/sec. 116 117 Actually, the idea is close to VJ's one, only timestamp cache is 118 held not per host, but per port pair and TW bucket is used as state 119 holder. 120 121 If TW bucket has been already destroyed we fall back to VJ's scheme 122 and use initial timestamp retrieved from peer table. 123 */ 124 if (tcptw->tw_ts_recent_stamp && 125 (!twp || (sysctl_tcp_tw_reuse && 126 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { 127 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; 128 if (tp->write_seq == 0) 129 tp->write_seq = 1; 130 tp->rx_opt.ts_recent = tcptw->tw_ts_recent; 131 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; 132 sock_hold(sktw); 133 return 1; 134 } 135 136 return 0; 137 } 138 EXPORT_SYMBOL_GPL(tcp_twsk_unique); 139 140 /* This will initiate an outgoing connection. */ 141 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 142 { 143 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; 144 struct inet_sock *inet = inet_sk(sk); 145 struct tcp_sock *tp = tcp_sk(sk); 146 __be16 orig_sport, orig_dport; 147 __be32 daddr, nexthop; 148 struct flowi4 *fl4; 149 struct rtable *rt; 150 int err; 151 struct ip_options_rcu *inet_opt; 152 153 if (addr_len < sizeof(struct sockaddr_in)) 154 return -EINVAL; 155 156 if (usin->sin_family != AF_INET) 157 return -EAFNOSUPPORT; 158 159 nexthop = daddr = usin->sin_addr.s_addr; 160 inet_opt = rcu_dereference_protected(inet->inet_opt, 161 sock_owned_by_user(sk)); 162 if (inet_opt && inet_opt->opt.srr) { 163 if (!daddr) 164 return -EINVAL; 165 nexthop = inet_opt->opt.faddr; 166 } 167 168 orig_sport = inet->inet_sport; 169 orig_dport = usin->sin_port; 170 fl4 = &inet->cork.fl.u.ip4; 171 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, 172 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, 173 IPPROTO_TCP, 174 orig_sport, orig_dport, sk); 175 if (IS_ERR(rt)) { 176 err = PTR_ERR(rt); 177 if (err == -ENETUNREACH) 178 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); 179 return err; 180 } 181 182 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { 183 ip_rt_put(rt); 184 return -ENETUNREACH; 185 } 186 187 if (!inet_opt || !inet_opt->opt.srr) 188 daddr = fl4->daddr; 189 190 if (!inet->inet_saddr) 191 inet->inet_saddr = fl4->saddr; 192 sk_rcv_saddr_set(sk, inet->inet_saddr); 193 194 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { 195 /* Reset inherited state */ 196 tp->rx_opt.ts_recent = 0; 197 tp->rx_opt.ts_recent_stamp = 0; 198 if (likely(!tp->repair)) 199 tp->write_seq = 0; 200 } 201 202 if (tcp_death_row.sysctl_tw_recycle && 203 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) 204 tcp_fetch_timewait_stamp(sk, &rt->dst); 205 206 inet->inet_dport = usin->sin_port; 207 sk_daddr_set(sk, daddr); 208 209 inet_csk(sk)->icsk_ext_hdr_len = 0; 210 if (inet_opt) 211 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; 212 213 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; 214 215 /* Socket identity is still unknown (sport may be zero). 216 * However we set state to SYN-SENT and not releasing socket 217 * lock select source port, enter ourselves into the hash tables and 218 * complete initialization after this. 219 */ 220 tcp_set_state(sk, TCP_SYN_SENT); 221 err = inet_hash_connect(&tcp_death_row, sk); 222 if (err) 223 goto failure; 224 225 sk_set_txhash(sk); 226 227 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, 228 inet->inet_sport, inet->inet_dport, sk); 229 if (IS_ERR(rt)) { 230 err = PTR_ERR(rt); 231 rt = NULL; 232 goto failure; 233 } 234 /* OK, now commit destination to socket. */ 235 sk->sk_gso_type = SKB_GSO_TCPV4; 236 sk_setup_caps(sk, &rt->dst); 237 238 if (!tp->write_seq && likely(!tp->repair)) 239 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, 240 inet->inet_daddr, 241 inet->inet_sport, 242 usin->sin_port); 243 244 inet->inet_id = tp->write_seq ^ jiffies; 245 246 err = tcp_connect(sk); 247 248 rt = NULL; 249 if (err) 250 goto failure; 251 252 return 0; 253 254 failure: 255 /* 256 * This unhashes the socket and releases the local port, 257 * if necessary. 258 */ 259 tcp_set_state(sk, TCP_CLOSE); 260 ip_rt_put(rt); 261 sk->sk_route_caps = 0; 262 inet->inet_dport = 0; 263 return err; 264 } 265 EXPORT_SYMBOL(tcp_v4_connect); 266 267 /* 268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. 269 * It can be called through tcp_release_cb() if socket was owned by user 270 * at the time tcp_v4_err() was called to handle ICMP message. 271 */ 272 void tcp_v4_mtu_reduced(struct sock *sk) 273 { 274 struct dst_entry *dst; 275 struct inet_sock *inet = inet_sk(sk); 276 u32 mtu = tcp_sk(sk)->mtu_info; 277 278 dst = inet_csk_update_pmtu(sk, mtu); 279 if (!dst) 280 return; 281 282 /* Something is about to be wrong... Remember soft error 283 * for the case, if this connection will not able to recover. 284 */ 285 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) 286 sk->sk_err_soft = EMSGSIZE; 287 288 mtu = dst_mtu(dst); 289 290 if (inet->pmtudisc != IP_PMTUDISC_DONT && 291 ip_sk_accept_pmtu(sk) && 292 inet_csk(sk)->icsk_pmtu_cookie > mtu) { 293 tcp_sync_mss(sk, mtu); 294 295 /* Resend the TCP packet because it's 296 * clear that the old packet has been 297 * dropped. This is the new "fast" path mtu 298 * discovery. 299 */ 300 tcp_simple_retransmit(sk); 301 } /* else let the usual retransmit timer handle it */ 302 } 303 EXPORT_SYMBOL(tcp_v4_mtu_reduced); 304 305 static void do_redirect(struct sk_buff *skb, struct sock *sk) 306 { 307 struct dst_entry *dst = __sk_dst_check(sk, 0); 308 309 if (dst) 310 dst->ops->redirect(dst, sk, skb); 311 } 312 313 314 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */ 315 void tcp_req_err(struct sock *sk, u32 seq) 316 { 317 struct request_sock *req = inet_reqsk(sk); 318 struct net *net = sock_net(sk); 319 320 /* ICMPs are not backlogged, hence we cannot get 321 * an established socket here. 322 */ 323 WARN_ON(req->sk); 324 325 if (seq != tcp_rsk(req)->snt_isn) { 326 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 327 } else { 328 /* 329 * Still in SYN_RECV, just remove it silently. 330 * There is no good way to pass the error to the newly 331 * created socket, and POSIX does not want network 332 * errors returned from accept(). 333 */ 334 inet_csk_reqsk_queue_drop(req->rsk_listener, req); 335 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS); 336 } 337 reqsk_put(req); 338 } 339 EXPORT_SYMBOL(tcp_req_err); 340 341 /* 342 * This routine is called by the ICMP module when it gets some 343 * sort of error condition. If err < 0 then the socket should 344 * be closed and the error returned to the user. If err > 0 345 * it's just the icmp type << 8 | icmp code. After adjustment 346 * header points to the first 8 bytes of the tcp header. We need 347 * to find the appropriate port. 348 * 349 * The locking strategy used here is very "optimistic". When 350 * someone else accesses the socket the ICMP is just dropped 351 * and for some paths there is no check at all. 352 * A more general error queue to queue errors for later handling 353 * is probably better. 354 * 355 */ 356 357 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) 358 { 359 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; 360 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); 361 struct inet_connection_sock *icsk; 362 struct tcp_sock *tp; 363 struct inet_sock *inet; 364 const int type = icmp_hdr(icmp_skb)->type; 365 const int code = icmp_hdr(icmp_skb)->code; 366 struct sock *sk; 367 struct sk_buff *skb; 368 struct request_sock *fastopen; 369 __u32 seq, snd_una; 370 __u32 remaining; 371 int err; 372 struct net *net = dev_net(icmp_skb->dev); 373 374 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr, 375 th->dest, iph->saddr, ntohs(th->source), 376 inet_iif(icmp_skb)); 377 if (!sk) { 378 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 379 return; 380 } 381 if (sk->sk_state == TCP_TIME_WAIT) { 382 inet_twsk_put(inet_twsk(sk)); 383 return; 384 } 385 seq = ntohl(th->seq); 386 if (sk->sk_state == TCP_NEW_SYN_RECV) 387 return tcp_req_err(sk, seq); 388 389 bh_lock_sock(sk); 390 /* If too many ICMPs get dropped on busy 391 * servers this needs to be solved differently. 392 * We do take care of PMTU discovery (RFC1191) special case : 393 * we can receive locally generated ICMP messages while socket is held. 394 */ 395 if (sock_owned_by_user(sk)) { 396 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) 397 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); 398 } 399 if (sk->sk_state == TCP_CLOSE) 400 goto out; 401 402 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { 403 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); 404 goto out; 405 } 406 407 icsk = inet_csk(sk); 408 tp = tcp_sk(sk); 409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ 410 fastopen = tp->fastopen_rsk; 411 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; 412 if (sk->sk_state != TCP_LISTEN && 413 !between(seq, snd_una, tp->snd_nxt)) { 414 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 415 goto out; 416 } 417 418 switch (type) { 419 case ICMP_REDIRECT: 420 do_redirect(icmp_skb, sk); 421 goto out; 422 case ICMP_SOURCE_QUENCH: 423 /* Just silently ignore these. */ 424 goto out; 425 case ICMP_PARAMETERPROB: 426 err = EPROTO; 427 break; 428 case ICMP_DEST_UNREACH: 429 if (code > NR_ICMP_UNREACH) 430 goto out; 431 432 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ 433 /* We are not interested in TCP_LISTEN and open_requests 434 * (SYN-ACKs send out by Linux are always <576bytes so 435 * they should go through unfragmented). 436 */ 437 if (sk->sk_state == TCP_LISTEN) 438 goto out; 439 440 tp->mtu_info = info; 441 if (!sock_owned_by_user(sk)) { 442 tcp_v4_mtu_reduced(sk); 443 } else { 444 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags)) 445 sock_hold(sk); 446 } 447 goto out; 448 } 449 450 err = icmp_err_convert[code].errno; 451 /* check if icmp_skb allows revert of backoff 452 * (see draft-zimmermann-tcp-lcd) */ 453 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) 454 break; 455 if (seq != tp->snd_una || !icsk->icsk_retransmits || 456 !icsk->icsk_backoff || fastopen) 457 break; 458 459 if (sock_owned_by_user(sk)) 460 break; 461 462 icsk->icsk_backoff--; 463 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : 464 TCP_TIMEOUT_INIT; 465 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX); 466 467 skb = tcp_write_queue_head(sk); 468 BUG_ON(!skb); 469 470 remaining = icsk->icsk_rto - 471 min(icsk->icsk_rto, 472 tcp_time_stamp - tcp_skb_timestamp(skb)); 473 474 if (remaining) { 475 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, 476 remaining, TCP_RTO_MAX); 477 } else { 478 /* RTO revert clocked out retransmission. 479 * Will retransmit now */ 480 tcp_retransmit_timer(sk); 481 } 482 483 break; 484 case ICMP_TIME_EXCEEDED: 485 err = EHOSTUNREACH; 486 break; 487 default: 488 goto out; 489 } 490 491 switch (sk->sk_state) { 492 case TCP_SYN_SENT: 493 case TCP_SYN_RECV: 494 /* Only in fast or simultaneous open. If a fast open socket is 495 * is already accepted it is treated as a connected one below. 496 */ 497 if (fastopen && !fastopen->sk) 498 break; 499 500 if (!sock_owned_by_user(sk)) { 501 sk->sk_err = err; 502 503 sk->sk_error_report(sk); 504 505 tcp_done(sk); 506 } else { 507 sk->sk_err_soft = err; 508 } 509 goto out; 510 } 511 512 /* If we've already connected we will keep trying 513 * until we time out, or the user gives up. 514 * 515 * rfc1122 4.2.3.9 allows to consider as hard errors 516 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, 517 * but it is obsoleted by pmtu discovery). 518 * 519 * Note, that in modern internet, where routing is unreliable 520 * and in each dark corner broken firewalls sit, sending random 521 * errors ordered by their masters even this two messages finally lose 522 * their original sense (even Linux sends invalid PORT_UNREACHs) 523 * 524 * Now we are in compliance with RFCs. 525 * --ANK (980905) 526 */ 527 528 inet = inet_sk(sk); 529 if (!sock_owned_by_user(sk) && inet->recverr) { 530 sk->sk_err = err; 531 sk->sk_error_report(sk); 532 } else { /* Only an error on timeout */ 533 sk->sk_err_soft = err; 534 } 535 536 out: 537 bh_unlock_sock(sk); 538 sock_put(sk); 539 } 540 541 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) 542 { 543 struct tcphdr *th = tcp_hdr(skb); 544 545 if (skb->ip_summed == CHECKSUM_PARTIAL) { 546 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); 547 skb->csum_start = skb_transport_header(skb) - skb->head; 548 skb->csum_offset = offsetof(struct tcphdr, check); 549 } else { 550 th->check = tcp_v4_check(skb->len, saddr, daddr, 551 csum_partial(th, 552 th->doff << 2, 553 skb->csum)); 554 } 555 } 556 557 /* This routine computes an IPv4 TCP checksum. */ 558 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) 559 { 560 const struct inet_sock *inet = inet_sk(sk); 561 562 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); 563 } 564 EXPORT_SYMBOL(tcp_v4_send_check); 565 566 /* 567 * This routine will send an RST to the other tcp. 568 * 569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) 570 * for reset. 571 * Answer: if a packet caused RST, it is not for a socket 572 * existing in our system, if it is matched to a socket, 573 * it is just duplicate segment or bug in other side's TCP. 574 * So that we build reply only basing on parameters 575 * arrived with segment. 576 * Exception: precedence violation. We do not implement it in any case. 577 */ 578 579 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb) 580 { 581 const struct tcphdr *th = tcp_hdr(skb); 582 struct { 583 struct tcphdr th; 584 #ifdef CONFIG_TCP_MD5SIG 585 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; 586 #endif 587 } rep; 588 struct ip_reply_arg arg; 589 #ifdef CONFIG_TCP_MD5SIG 590 struct tcp_md5sig_key *key; 591 const __u8 *hash_location = NULL; 592 unsigned char newhash[16]; 593 int genhash; 594 struct sock *sk1 = NULL; 595 #endif 596 struct net *net; 597 598 /* Never send a reset in response to a reset. */ 599 if (th->rst) 600 return; 601 602 /* If sk not NULL, it means we did a successful lookup and incoming 603 * route had to be correct. prequeue might have dropped our dst. 604 */ 605 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL) 606 return; 607 608 /* Swap the send and the receive. */ 609 memset(&rep, 0, sizeof(rep)); 610 rep.th.dest = th->source; 611 rep.th.source = th->dest; 612 rep.th.doff = sizeof(struct tcphdr) / 4; 613 rep.th.rst = 1; 614 615 if (th->ack) { 616 rep.th.seq = th->ack_seq; 617 } else { 618 rep.th.ack = 1; 619 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + 620 skb->len - (th->doff << 2)); 621 } 622 623 memset(&arg, 0, sizeof(arg)); 624 arg.iov[0].iov_base = (unsigned char *)&rep; 625 arg.iov[0].iov_len = sizeof(rep.th); 626 627 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); 628 #ifdef CONFIG_TCP_MD5SIG 629 hash_location = tcp_parse_md5sig_option(th); 630 if (!sk && hash_location) { 631 /* 632 * active side is lost. Try to find listening socket through 633 * source port, and then find md5 key through listening socket. 634 * we are not loose security here: 635 * Incoming packet is checked with md5 hash with finding key, 636 * no RST generated if md5 hash doesn't match. 637 */ 638 sk1 = __inet_lookup_listener(net, 639 &tcp_hashinfo, ip_hdr(skb)->saddr, 640 th->source, ip_hdr(skb)->daddr, 641 ntohs(th->source), inet_iif(skb)); 642 /* don't send rst if it can't find key */ 643 if (!sk1) 644 return; 645 rcu_read_lock(); 646 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) 647 &ip_hdr(skb)->saddr, AF_INET); 648 if (!key) 649 goto release_sk1; 650 651 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb); 652 if (genhash || memcmp(hash_location, newhash, 16) != 0) 653 goto release_sk1; 654 } else { 655 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *) 656 &ip_hdr(skb)->saddr, 657 AF_INET) : NULL; 658 } 659 660 if (key) { 661 rep.opt[0] = htonl((TCPOPT_NOP << 24) | 662 (TCPOPT_NOP << 16) | 663 (TCPOPT_MD5SIG << 8) | 664 TCPOLEN_MD5SIG); 665 /* Update length and the length the header thinks exists */ 666 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; 667 rep.th.doff = arg.iov[0].iov_len / 4; 668 669 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], 670 key, ip_hdr(skb)->saddr, 671 ip_hdr(skb)->daddr, &rep.th); 672 } 673 #endif 674 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, 675 ip_hdr(skb)->saddr, /* XXX */ 676 arg.iov[0].iov_len, IPPROTO_TCP, 0); 677 arg.csumoffset = offsetof(struct tcphdr, check) / 2; 678 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; 679 /* When socket is gone, all binding information is lost. 680 * routing might fail in this case. No choice here, if we choose to force 681 * input interface, we will misroute in case of asymmetric route. 682 */ 683 if (sk) 684 arg.bound_dev_if = sk->sk_bound_dev_if; 685 686 arg.tos = ip_hdr(skb)->tos; 687 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk), 688 skb, &TCP_SKB_CB(skb)->header.h4.opt, 689 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, 690 &arg, arg.iov[0].iov_len); 691 692 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); 693 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); 694 695 #ifdef CONFIG_TCP_MD5SIG 696 release_sk1: 697 if (sk1) { 698 rcu_read_unlock(); 699 sock_put(sk1); 700 } 701 #endif 702 } 703 704 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states 705 outside socket context is ugly, certainly. What can I do? 706 */ 707 708 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, 709 u32 win, u32 tsval, u32 tsecr, int oif, 710 struct tcp_md5sig_key *key, 711 int reply_flags, u8 tos) 712 { 713 const struct tcphdr *th = tcp_hdr(skb); 714 struct { 715 struct tcphdr th; 716 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) 717 #ifdef CONFIG_TCP_MD5SIG 718 + (TCPOLEN_MD5SIG_ALIGNED >> 2) 719 #endif 720 ]; 721 } rep; 722 struct ip_reply_arg arg; 723 struct net *net = dev_net(skb_dst(skb)->dev); 724 725 memset(&rep.th, 0, sizeof(struct tcphdr)); 726 memset(&arg, 0, sizeof(arg)); 727 728 arg.iov[0].iov_base = (unsigned char *)&rep; 729 arg.iov[0].iov_len = sizeof(rep.th); 730 if (tsecr) { 731 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 732 (TCPOPT_TIMESTAMP << 8) | 733 TCPOLEN_TIMESTAMP); 734 rep.opt[1] = htonl(tsval); 735 rep.opt[2] = htonl(tsecr); 736 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; 737 } 738 739 /* Swap the send and the receive. */ 740 rep.th.dest = th->source; 741 rep.th.source = th->dest; 742 rep.th.doff = arg.iov[0].iov_len / 4; 743 rep.th.seq = htonl(seq); 744 rep.th.ack_seq = htonl(ack); 745 rep.th.ack = 1; 746 rep.th.window = htons(win); 747 748 #ifdef CONFIG_TCP_MD5SIG 749 if (key) { 750 int offset = (tsecr) ? 3 : 0; 751 752 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | 753 (TCPOPT_NOP << 16) | 754 (TCPOPT_MD5SIG << 8) | 755 TCPOLEN_MD5SIG); 756 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; 757 rep.th.doff = arg.iov[0].iov_len/4; 758 759 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], 760 key, ip_hdr(skb)->saddr, 761 ip_hdr(skb)->daddr, &rep.th); 762 } 763 #endif 764 arg.flags = reply_flags; 765 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, 766 ip_hdr(skb)->saddr, /* XXX */ 767 arg.iov[0].iov_len, IPPROTO_TCP, 0); 768 arg.csumoffset = offsetof(struct tcphdr, check) / 2; 769 if (oif) 770 arg.bound_dev_if = oif; 771 arg.tos = tos; 772 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk), 773 skb, &TCP_SKB_CB(skb)->header.h4.opt, 774 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, 775 &arg, arg.iov[0].iov_len); 776 777 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); 778 } 779 780 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) 781 { 782 struct inet_timewait_sock *tw = inet_twsk(sk); 783 struct tcp_timewait_sock *tcptw = tcp_twsk(sk); 784 785 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, 786 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, 787 tcp_time_stamp + tcptw->tw_ts_offset, 788 tcptw->tw_ts_recent, 789 tw->tw_bound_dev_if, 790 tcp_twsk_md5_key(tcptw), 791 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, 792 tw->tw_tos 793 ); 794 795 inet_twsk_put(tw); 796 } 797 798 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 799 struct request_sock *req) 800 { 801 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV 802 * sk->sk_state == TCP_SYN_RECV -> for Fast Open. 803 */ 804 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ? 805 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, 806 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd, 807 tcp_time_stamp, 808 req->ts_recent, 809 0, 810 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, 811 AF_INET), 812 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, 813 ip_hdr(skb)->tos); 814 } 815 816 /* 817 * Send a SYN-ACK after having received a SYN. 818 * This still operates on a request_sock only, not on a big 819 * socket. 820 */ 821 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, 822 struct flowi *fl, 823 struct request_sock *req, 824 struct tcp_fastopen_cookie *foc, 825 bool attach_req) 826 { 827 const struct inet_request_sock *ireq = inet_rsk(req); 828 struct flowi4 fl4; 829 int err = -1; 830 struct sk_buff *skb; 831 832 /* First, grab a route. */ 833 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) 834 return -1; 835 836 skb = tcp_make_synack(sk, dst, req, foc, attach_req); 837 838 if (skb) { 839 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); 840 841 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, 842 ireq->ir_rmt_addr, 843 ireq->opt); 844 err = net_xmit_eval(err); 845 } 846 847 return err; 848 } 849 850 /* 851 * IPv4 request_sock destructor. 852 */ 853 static void tcp_v4_reqsk_destructor(struct request_sock *req) 854 { 855 kfree(inet_rsk(req)->opt); 856 } 857 858 859 #ifdef CONFIG_TCP_MD5SIG 860 /* 861 * RFC2385 MD5 checksumming requires a mapping of 862 * IP address->MD5 Key. 863 * We need to maintain these in the sk structure. 864 */ 865 866 /* Find the Key structure for an address. */ 867 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 868 const union tcp_md5_addr *addr, 869 int family) 870 { 871 const struct tcp_sock *tp = tcp_sk(sk); 872 struct tcp_md5sig_key *key; 873 unsigned int size = sizeof(struct in_addr); 874 const struct tcp_md5sig_info *md5sig; 875 876 /* caller either holds rcu_read_lock() or socket lock */ 877 md5sig = rcu_dereference_check(tp->md5sig_info, 878 sock_owned_by_user(sk) || 879 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock)); 880 if (!md5sig) 881 return NULL; 882 #if IS_ENABLED(CONFIG_IPV6) 883 if (family == AF_INET6) 884 size = sizeof(struct in6_addr); 885 #endif 886 hlist_for_each_entry_rcu(key, &md5sig->head, node) { 887 if (key->family != family) 888 continue; 889 if (!memcmp(&key->addr, addr, size)) 890 return key; 891 } 892 return NULL; 893 } 894 EXPORT_SYMBOL(tcp_md5_do_lookup); 895 896 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 897 const struct sock *addr_sk) 898 { 899 const union tcp_md5_addr *addr; 900 901 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr; 902 return tcp_md5_do_lookup(sk, addr, AF_INET); 903 } 904 EXPORT_SYMBOL(tcp_v4_md5_lookup); 905 906 /* This can be called on a newly created socket, from other files */ 907 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 908 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) 909 { 910 /* Add Key to the list */ 911 struct tcp_md5sig_key *key; 912 struct tcp_sock *tp = tcp_sk(sk); 913 struct tcp_md5sig_info *md5sig; 914 915 key = tcp_md5_do_lookup(sk, addr, family); 916 if (key) { 917 /* Pre-existing entry - just update that one. */ 918 memcpy(key->key, newkey, newkeylen); 919 key->keylen = newkeylen; 920 return 0; 921 } 922 923 md5sig = rcu_dereference_protected(tp->md5sig_info, 924 sock_owned_by_user(sk)); 925 if (!md5sig) { 926 md5sig = kmalloc(sizeof(*md5sig), gfp); 927 if (!md5sig) 928 return -ENOMEM; 929 930 sk_nocaps_add(sk, NETIF_F_GSO_MASK); 931 INIT_HLIST_HEAD(&md5sig->head); 932 rcu_assign_pointer(tp->md5sig_info, md5sig); 933 } 934 935 key = sock_kmalloc(sk, sizeof(*key), gfp); 936 if (!key) 937 return -ENOMEM; 938 if (!tcp_alloc_md5sig_pool()) { 939 sock_kfree_s(sk, key, sizeof(*key)); 940 return -ENOMEM; 941 } 942 943 memcpy(key->key, newkey, newkeylen); 944 key->keylen = newkeylen; 945 key->family = family; 946 memcpy(&key->addr, addr, 947 (family == AF_INET6) ? sizeof(struct in6_addr) : 948 sizeof(struct in_addr)); 949 hlist_add_head_rcu(&key->node, &md5sig->head); 950 return 0; 951 } 952 EXPORT_SYMBOL(tcp_md5_do_add); 953 954 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) 955 { 956 struct tcp_md5sig_key *key; 957 958 key = tcp_md5_do_lookup(sk, addr, family); 959 if (!key) 960 return -ENOENT; 961 hlist_del_rcu(&key->node); 962 atomic_sub(sizeof(*key), &sk->sk_omem_alloc); 963 kfree_rcu(key, rcu); 964 return 0; 965 } 966 EXPORT_SYMBOL(tcp_md5_do_del); 967 968 static void tcp_clear_md5_list(struct sock *sk) 969 { 970 struct tcp_sock *tp = tcp_sk(sk); 971 struct tcp_md5sig_key *key; 972 struct hlist_node *n; 973 struct tcp_md5sig_info *md5sig; 974 975 md5sig = rcu_dereference_protected(tp->md5sig_info, 1); 976 977 hlist_for_each_entry_safe(key, n, &md5sig->head, node) { 978 hlist_del_rcu(&key->node); 979 atomic_sub(sizeof(*key), &sk->sk_omem_alloc); 980 kfree_rcu(key, rcu); 981 } 982 } 983 984 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, 985 int optlen) 986 { 987 struct tcp_md5sig cmd; 988 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; 989 990 if (optlen < sizeof(cmd)) 991 return -EINVAL; 992 993 if (copy_from_user(&cmd, optval, sizeof(cmd))) 994 return -EFAULT; 995 996 if (sin->sin_family != AF_INET) 997 return -EINVAL; 998 999 if (!cmd.tcpm_keylen) 1000 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, 1001 AF_INET); 1002 1003 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 1004 return -EINVAL; 1005 1006 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, 1007 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, 1008 GFP_KERNEL); 1009 } 1010 1011 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, 1012 __be32 daddr, __be32 saddr, int nbytes) 1013 { 1014 struct tcp4_pseudohdr *bp; 1015 struct scatterlist sg; 1016 1017 bp = &hp->md5_blk.ip4; 1018 1019 /* 1020 * 1. the TCP pseudo-header (in the order: source IP address, 1021 * destination IP address, zero-padded protocol number, and 1022 * segment length) 1023 */ 1024 bp->saddr = saddr; 1025 bp->daddr = daddr; 1026 bp->pad = 0; 1027 bp->protocol = IPPROTO_TCP; 1028 bp->len = cpu_to_be16(nbytes); 1029 1030 sg_init_one(&sg, bp, sizeof(*bp)); 1031 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); 1032 } 1033 1034 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 1035 __be32 daddr, __be32 saddr, const struct tcphdr *th) 1036 { 1037 struct tcp_md5sig_pool *hp; 1038 struct hash_desc *desc; 1039 1040 hp = tcp_get_md5sig_pool(); 1041 if (!hp) 1042 goto clear_hash_noput; 1043 desc = &hp->md5_desc; 1044 1045 if (crypto_hash_init(desc)) 1046 goto clear_hash; 1047 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) 1048 goto clear_hash; 1049 if (tcp_md5_hash_header(hp, th)) 1050 goto clear_hash; 1051 if (tcp_md5_hash_key(hp, key)) 1052 goto clear_hash; 1053 if (crypto_hash_final(desc, md5_hash)) 1054 goto clear_hash; 1055 1056 tcp_put_md5sig_pool(); 1057 return 0; 1058 1059 clear_hash: 1060 tcp_put_md5sig_pool(); 1061 clear_hash_noput: 1062 memset(md5_hash, 0, 16); 1063 return 1; 1064 } 1065 1066 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1067 const struct sock *sk, 1068 const struct sk_buff *skb) 1069 { 1070 struct tcp_md5sig_pool *hp; 1071 struct hash_desc *desc; 1072 const struct tcphdr *th = tcp_hdr(skb); 1073 __be32 saddr, daddr; 1074 1075 if (sk) { /* valid for establish/request sockets */ 1076 saddr = sk->sk_rcv_saddr; 1077 daddr = sk->sk_daddr; 1078 } else { 1079 const struct iphdr *iph = ip_hdr(skb); 1080 saddr = iph->saddr; 1081 daddr = iph->daddr; 1082 } 1083 1084 hp = tcp_get_md5sig_pool(); 1085 if (!hp) 1086 goto clear_hash_noput; 1087 desc = &hp->md5_desc; 1088 1089 if (crypto_hash_init(desc)) 1090 goto clear_hash; 1091 1092 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) 1093 goto clear_hash; 1094 if (tcp_md5_hash_header(hp, th)) 1095 goto clear_hash; 1096 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 1097 goto clear_hash; 1098 if (tcp_md5_hash_key(hp, key)) 1099 goto clear_hash; 1100 if (crypto_hash_final(desc, md5_hash)) 1101 goto clear_hash; 1102 1103 tcp_put_md5sig_pool(); 1104 return 0; 1105 1106 clear_hash: 1107 tcp_put_md5sig_pool(); 1108 clear_hash_noput: 1109 memset(md5_hash, 0, 16); 1110 return 1; 1111 } 1112 EXPORT_SYMBOL(tcp_v4_md5_hash_skb); 1113 1114 #endif 1115 1116 /* Called with rcu_read_lock() */ 1117 static bool tcp_v4_inbound_md5_hash(const struct sock *sk, 1118 const struct sk_buff *skb) 1119 { 1120 #ifdef CONFIG_TCP_MD5SIG 1121 /* 1122 * This gets called for each TCP segment that arrives 1123 * so we want to be efficient. 1124 * We have 3 drop cases: 1125 * o No MD5 hash and one expected. 1126 * o MD5 hash and we're not expecting one. 1127 * o MD5 hash and its wrong. 1128 */ 1129 const __u8 *hash_location = NULL; 1130 struct tcp_md5sig_key *hash_expected; 1131 const struct iphdr *iph = ip_hdr(skb); 1132 const struct tcphdr *th = tcp_hdr(skb); 1133 int genhash; 1134 unsigned char newhash[16]; 1135 1136 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, 1137 AF_INET); 1138 hash_location = tcp_parse_md5sig_option(th); 1139 1140 /* We've parsed the options - do we have a hash? */ 1141 if (!hash_expected && !hash_location) 1142 return false; 1143 1144 if (hash_expected && !hash_location) { 1145 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 1146 return true; 1147 } 1148 1149 if (!hash_expected && hash_location) { 1150 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); 1151 return true; 1152 } 1153 1154 /* Okay, so this is hash_expected and hash_location - 1155 * so we need to calculate the checksum. 1156 */ 1157 genhash = tcp_v4_md5_hash_skb(newhash, 1158 hash_expected, 1159 NULL, skb); 1160 1161 if (genhash || memcmp(hash_location, newhash, 16) != 0) { 1162 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", 1163 &iph->saddr, ntohs(th->source), 1164 &iph->daddr, ntohs(th->dest), 1165 genhash ? " tcp_v4_calc_md5_hash failed" 1166 : ""); 1167 return true; 1168 } 1169 return false; 1170 #endif 1171 return false; 1172 } 1173 1174 static void tcp_v4_init_req(struct request_sock *req, 1175 const struct sock *sk_listener, 1176 struct sk_buff *skb) 1177 { 1178 struct inet_request_sock *ireq = inet_rsk(req); 1179 1180 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); 1181 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); 1182 ireq->no_srccheck = inet_sk(sk_listener)->transparent; 1183 ireq->opt = tcp_v4_save_options(skb); 1184 } 1185 1186 static struct dst_entry *tcp_v4_route_req(const struct sock *sk, 1187 struct flowi *fl, 1188 const struct request_sock *req, 1189 bool *strict) 1190 { 1191 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req); 1192 1193 if (strict) { 1194 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr) 1195 *strict = true; 1196 else 1197 *strict = false; 1198 } 1199 1200 return dst; 1201 } 1202 1203 struct request_sock_ops tcp_request_sock_ops __read_mostly = { 1204 .family = PF_INET, 1205 .obj_size = sizeof(struct tcp_request_sock), 1206 .rtx_syn_ack = tcp_rtx_synack, 1207 .send_ack = tcp_v4_reqsk_send_ack, 1208 .destructor = tcp_v4_reqsk_destructor, 1209 .send_reset = tcp_v4_send_reset, 1210 .syn_ack_timeout = tcp_syn_ack_timeout, 1211 }; 1212 1213 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { 1214 .mss_clamp = TCP_MSS_DEFAULT, 1215 #ifdef CONFIG_TCP_MD5SIG 1216 .req_md5_lookup = tcp_v4_md5_lookup, 1217 .calc_md5_hash = tcp_v4_md5_hash_skb, 1218 #endif 1219 .init_req = tcp_v4_init_req, 1220 #ifdef CONFIG_SYN_COOKIES 1221 .cookie_init_seq = cookie_v4_init_sequence, 1222 #endif 1223 .route_req = tcp_v4_route_req, 1224 .init_seq = tcp_v4_init_sequence, 1225 .send_synack = tcp_v4_send_synack, 1226 }; 1227 1228 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) 1229 { 1230 /* Never answer to SYNs send to broadcast or multicast */ 1231 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 1232 goto drop; 1233 1234 return tcp_conn_request(&tcp_request_sock_ops, 1235 &tcp_request_sock_ipv4_ops, sk, skb); 1236 1237 drop: 1238 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 1239 return 0; 1240 } 1241 EXPORT_SYMBOL(tcp_v4_conn_request); 1242 1243 1244 /* 1245 * The three way handshake has completed - we got a valid synack - 1246 * now create the new socket. 1247 */ 1248 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 1249 struct request_sock *req, 1250 struct dst_entry *dst, 1251 struct request_sock *req_unhash, 1252 bool *own_req) 1253 { 1254 struct inet_request_sock *ireq; 1255 struct inet_sock *newinet; 1256 struct tcp_sock *newtp; 1257 struct sock *newsk; 1258 #ifdef CONFIG_TCP_MD5SIG 1259 struct tcp_md5sig_key *key; 1260 #endif 1261 struct ip_options_rcu *inet_opt; 1262 1263 if (sk_acceptq_is_full(sk)) 1264 goto exit_overflow; 1265 1266 newsk = tcp_create_openreq_child(sk, req, skb); 1267 if (!newsk) 1268 goto exit_nonewsk; 1269 1270 newsk->sk_gso_type = SKB_GSO_TCPV4; 1271 inet_sk_rx_dst_set(newsk, skb); 1272 1273 newtp = tcp_sk(newsk); 1274 newinet = inet_sk(newsk); 1275 ireq = inet_rsk(req); 1276 sk_daddr_set(newsk, ireq->ir_rmt_addr); 1277 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); 1278 newinet->inet_saddr = ireq->ir_loc_addr; 1279 inet_opt = ireq->opt; 1280 rcu_assign_pointer(newinet->inet_opt, inet_opt); 1281 ireq->opt = NULL; 1282 newinet->mc_index = inet_iif(skb); 1283 newinet->mc_ttl = ip_hdr(skb)->ttl; 1284 newinet->rcv_tos = ip_hdr(skb)->tos; 1285 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1286 if (inet_opt) 1287 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; 1288 newinet->inet_id = newtp->write_seq ^ jiffies; 1289 1290 if (!dst) { 1291 dst = inet_csk_route_child_sock(sk, newsk, req); 1292 if (!dst) 1293 goto put_and_exit; 1294 } else { 1295 /* syncookie case : see end of cookie_v4_check() */ 1296 } 1297 sk_setup_caps(newsk, dst); 1298 1299 tcp_ca_openreq_child(newsk, dst); 1300 1301 tcp_sync_mss(newsk, dst_mtu(dst)); 1302 newtp->advmss = dst_metric_advmss(dst); 1303 if (tcp_sk(sk)->rx_opt.user_mss && 1304 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) 1305 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; 1306 1307 tcp_initialize_rcv_mss(newsk); 1308 1309 #ifdef CONFIG_TCP_MD5SIG 1310 /* Copy over the MD5 key from the original socket */ 1311 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, 1312 AF_INET); 1313 if (key) { 1314 /* 1315 * We're using one, so create a matching key 1316 * on the newsk structure. If we fail to get 1317 * memory, then we end up not copying the key 1318 * across. Shucks. 1319 */ 1320 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, 1321 AF_INET, key->key, key->keylen, GFP_ATOMIC); 1322 sk_nocaps_add(newsk, NETIF_F_GSO_MASK); 1323 } 1324 #endif 1325 1326 if (__inet_inherit_port(sk, newsk) < 0) 1327 goto put_and_exit; 1328 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); 1329 if (*own_req) 1330 tcp_move_syn(newtp, req); 1331 1332 return newsk; 1333 1334 exit_overflow: 1335 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1336 exit_nonewsk: 1337 dst_release(dst); 1338 exit: 1339 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 1340 return NULL; 1341 put_and_exit: 1342 inet_csk_prepare_forced_close(newsk); 1343 tcp_done(newsk); 1344 goto exit; 1345 } 1346 EXPORT_SYMBOL(tcp_v4_syn_recv_sock); 1347 1348 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb) 1349 { 1350 #ifdef CONFIG_SYN_COOKIES 1351 const struct tcphdr *th = tcp_hdr(skb); 1352 1353 if (!th->syn) 1354 sk = cookie_v4_check(sk, skb); 1355 #endif 1356 return sk; 1357 } 1358 1359 /* The socket must have it's spinlock held when we get 1360 * here, unless it is a TCP_LISTEN socket. 1361 * 1362 * We have a potential double-lock case here, so even when 1363 * doing backlog processing we use the BH locking scheme. 1364 * This is because we cannot sleep with the original spinlock 1365 * held. 1366 */ 1367 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) 1368 { 1369 struct sock *rsk; 1370 1371 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1372 struct dst_entry *dst = sk->sk_rx_dst; 1373 1374 sock_rps_save_rxhash(sk, skb); 1375 sk_mark_napi_id(sk, skb); 1376 if (dst) { 1377 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || 1378 !dst->ops->check(dst, 0)) { 1379 dst_release(dst); 1380 sk->sk_rx_dst = NULL; 1381 } 1382 } 1383 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); 1384 return 0; 1385 } 1386 1387 if (tcp_checksum_complete(skb)) 1388 goto csum_err; 1389 1390 if (sk->sk_state == TCP_LISTEN) { 1391 struct sock *nsk = tcp_v4_cookie_check(sk, skb); 1392 1393 if (!nsk) 1394 goto discard; 1395 if (nsk != sk) { 1396 sock_rps_save_rxhash(nsk, skb); 1397 sk_mark_napi_id(nsk, skb); 1398 if (tcp_child_process(sk, nsk, skb)) { 1399 rsk = nsk; 1400 goto reset; 1401 } 1402 return 0; 1403 } 1404 } else 1405 sock_rps_save_rxhash(sk, skb); 1406 1407 if (tcp_rcv_state_process(sk, skb)) { 1408 rsk = sk; 1409 goto reset; 1410 } 1411 return 0; 1412 1413 reset: 1414 tcp_v4_send_reset(rsk, skb); 1415 discard: 1416 kfree_skb(skb); 1417 /* Be careful here. If this function gets more complicated and 1418 * gcc suffers from register pressure on the x86, sk (in %ebx) 1419 * might be destroyed here. This current version compiles correctly, 1420 * but you have been warned. 1421 */ 1422 return 0; 1423 1424 csum_err: 1425 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); 1426 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); 1427 goto discard; 1428 } 1429 EXPORT_SYMBOL(tcp_v4_do_rcv); 1430 1431 void tcp_v4_early_demux(struct sk_buff *skb) 1432 { 1433 const struct iphdr *iph; 1434 const struct tcphdr *th; 1435 struct sock *sk; 1436 1437 if (skb->pkt_type != PACKET_HOST) 1438 return; 1439 1440 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1441 return; 1442 1443 iph = ip_hdr(skb); 1444 th = tcp_hdr(skb); 1445 1446 if (th->doff < sizeof(struct tcphdr) / 4) 1447 return; 1448 1449 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, 1450 iph->saddr, th->source, 1451 iph->daddr, ntohs(th->dest), 1452 skb->skb_iif); 1453 if (sk) { 1454 skb->sk = sk; 1455 skb->destructor = sock_edemux; 1456 if (sk_fullsock(sk)) { 1457 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); 1458 1459 if (dst) 1460 dst = dst_check(dst, 0); 1461 if (dst && 1462 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) 1463 skb_dst_set_noref(skb, dst); 1464 } 1465 } 1466 } 1467 1468 /* Packet is added to VJ-style prequeue for processing in process 1469 * context, if a reader task is waiting. Apparently, this exciting 1470 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 1471 * failed somewhere. Latency? Burstiness? Well, at least now we will 1472 * see, why it failed. 8)8) --ANK 1473 * 1474 */ 1475 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) 1476 { 1477 struct tcp_sock *tp = tcp_sk(sk); 1478 1479 if (sysctl_tcp_low_latency || !tp->ucopy.task) 1480 return false; 1481 1482 if (skb->len <= tcp_hdrlen(skb) && 1483 skb_queue_len(&tp->ucopy.prequeue) == 0) 1484 return false; 1485 1486 /* Before escaping RCU protected region, we need to take care of skb 1487 * dst. Prequeue is only enabled for established sockets. 1488 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst 1489 * Instead of doing full sk_rx_dst validity here, let's perform 1490 * an optimistic check. 1491 */ 1492 if (likely(sk->sk_rx_dst)) 1493 skb_dst_drop(skb); 1494 else 1495 skb_dst_force(skb); 1496 1497 __skb_queue_tail(&tp->ucopy.prequeue, skb); 1498 tp->ucopy.memory += skb->truesize; 1499 if (tp->ucopy.memory > sk->sk_rcvbuf) { 1500 struct sk_buff *skb1; 1501 1502 BUG_ON(sock_owned_by_user(sk)); 1503 1504 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 1505 sk_backlog_rcv(sk, skb1); 1506 NET_INC_STATS_BH(sock_net(sk), 1507 LINUX_MIB_TCPPREQUEUEDROPPED); 1508 } 1509 1510 tp->ucopy.memory = 0; 1511 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 1512 wake_up_interruptible_sync_poll(sk_sleep(sk), 1513 POLLIN | POLLRDNORM | POLLRDBAND); 1514 if (!inet_csk_ack_scheduled(sk)) 1515 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 1516 (3 * tcp_rto_min(sk)) / 4, 1517 TCP_RTO_MAX); 1518 } 1519 return true; 1520 } 1521 EXPORT_SYMBOL(tcp_prequeue); 1522 1523 /* 1524 * From tcp_input.c 1525 */ 1526 1527 int tcp_v4_rcv(struct sk_buff *skb) 1528 { 1529 const struct iphdr *iph; 1530 const struct tcphdr *th; 1531 struct sock *sk; 1532 int ret; 1533 struct net *net = dev_net(skb->dev); 1534 1535 if (skb->pkt_type != PACKET_HOST) 1536 goto discard_it; 1537 1538 /* Count it even if it's bad */ 1539 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); 1540 1541 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1542 goto discard_it; 1543 1544 th = tcp_hdr(skb); 1545 1546 if (th->doff < sizeof(struct tcphdr) / 4) 1547 goto bad_packet; 1548 if (!pskb_may_pull(skb, th->doff * 4)) 1549 goto discard_it; 1550 1551 /* An explanation is required here, I think. 1552 * Packet length and doff are validated by header prediction, 1553 * provided case of th->doff==0 is eliminated. 1554 * So, we defer the checks. */ 1555 1556 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) 1557 goto csum_error; 1558 1559 th = tcp_hdr(skb); 1560 iph = ip_hdr(skb); 1561 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB() 1562 * barrier() makes sure compiler wont play fool^Waliasing games. 1563 */ 1564 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb), 1565 sizeof(struct inet_skb_parm)); 1566 barrier(); 1567 1568 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1569 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1570 skb->len - th->doff * 4); 1571 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1572 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1573 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1574 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); 1575 TCP_SKB_CB(skb)->sacked = 0; 1576 1577 lookup: 1578 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); 1579 if (!sk) 1580 goto no_tcp_socket; 1581 1582 process: 1583 if (sk->sk_state == TCP_TIME_WAIT) 1584 goto do_time_wait; 1585 1586 if (sk->sk_state == TCP_NEW_SYN_RECV) { 1587 struct request_sock *req = inet_reqsk(sk); 1588 struct sock *nsk = NULL; 1589 1590 sk = req->rsk_listener; 1591 if (tcp_v4_inbound_md5_hash(sk, skb)) 1592 goto discard_and_relse; 1593 if (likely(sk->sk_state == TCP_LISTEN)) { 1594 nsk = tcp_check_req(sk, skb, req, false); 1595 } else { 1596 inet_csk_reqsk_queue_drop_and_put(sk, req); 1597 goto lookup; 1598 } 1599 if (!nsk) { 1600 reqsk_put(req); 1601 goto discard_it; 1602 } 1603 if (nsk == sk) { 1604 sock_hold(sk); 1605 reqsk_put(req); 1606 } else if (tcp_child_process(sk, nsk, skb)) { 1607 tcp_v4_send_reset(nsk, skb); 1608 goto discard_it; 1609 } else { 1610 return 0; 1611 } 1612 } 1613 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { 1614 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); 1615 goto discard_and_relse; 1616 } 1617 1618 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1619 goto discard_and_relse; 1620 1621 if (tcp_v4_inbound_md5_hash(sk, skb)) 1622 goto discard_and_relse; 1623 1624 nf_reset(skb); 1625 1626 if (sk_filter(sk, skb)) 1627 goto discard_and_relse; 1628 1629 skb->dev = NULL; 1630 1631 if (sk->sk_state == TCP_LISTEN) { 1632 ret = tcp_v4_do_rcv(sk, skb); 1633 goto put_and_return; 1634 } 1635 1636 sk_incoming_cpu_update(sk); 1637 1638 bh_lock_sock_nested(sk); 1639 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1640 ret = 0; 1641 if (!sock_owned_by_user(sk)) { 1642 if (!tcp_prequeue(sk, skb)) 1643 ret = tcp_v4_do_rcv(sk, skb); 1644 } else if (unlikely(sk_add_backlog(sk, skb, 1645 sk->sk_rcvbuf + sk->sk_sndbuf))) { 1646 bh_unlock_sock(sk); 1647 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); 1648 goto discard_and_relse; 1649 } 1650 bh_unlock_sock(sk); 1651 1652 put_and_return: 1653 sock_put(sk); 1654 1655 return ret; 1656 1657 no_tcp_socket: 1658 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1659 goto discard_it; 1660 1661 if (tcp_checksum_complete(skb)) { 1662 csum_error: 1663 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS); 1664 bad_packet: 1665 TCP_INC_STATS_BH(net, TCP_MIB_INERRS); 1666 } else { 1667 tcp_v4_send_reset(NULL, skb); 1668 } 1669 1670 discard_it: 1671 /* Discard frame. */ 1672 kfree_skb(skb); 1673 return 0; 1674 1675 discard_and_relse: 1676 sock_put(sk); 1677 goto discard_it; 1678 1679 do_time_wait: 1680 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1681 inet_twsk_put(inet_twsk(sk)); 1682 goto discard_it; 1683 } 1684 1685 if (tcp_checksum_complete(skb)) { 1686 inet_twsk_put(inet_twsk(sk)); 1687 goto csum_error; 1688 } 1689 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1690 case TCP_TW_SYN: { 1691 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), 1692 &tcp_hashinfo, 1693 iph->saddr, th->source, 1694 iph->daddr, th->dest, 1695 inet_iif(skb)); 1696 if (sk2) { 1697 inet_twsk_deschedule_put(inet_twsk(sk)); 1698 sk = sk2; 1699 goto process; 1700 } 1701 /* Fall through to ACK */ 1702 } 1703 case TCP_TW_ACK: 1704 tcp_v4_timewait_ack(sk, skb); 1705 break; 1706 case TCP_TW_RST: 1707 goto no_tcp_socket; 1708 case TCP_TW_SUCCESS:; 1709 } 1710 goto discard_it; 1711 } 1712 1713 static struct timewait_sock_ops tcp_timewait_sock_ops = { 1714 .twsk_obj_size = sizeof(struct tcp_timewait_sock), 1715 .twsk_unique = tcp_twsk_unique, 1716 .twsk_destructor= tcp_twsk_destructor, 1717 }; 1718 1719 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) 1720 { 1721 struct dst_entry *dst = skb_dst(skb); 1722 1723 if (dst) { 1724 dst_hold(dst); 1725 sk->sk_rx_dst = dst; 1726 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; 1727 } 1728 } 1729 EXPORT_SYMBOL(inet_sk_rx_dst_set); 1730 1731 const struct inet_connection_sock_af_ops ipv4_specific = { 1732 .queue_xmit = ip_queue_xmit, 1733 .send_check = tcp_v4_send_check, 1734 .rebuild_header = inet_sk_rebuild_header, 1735 .sk_rx_dst_set = inet_sk_rx_dst_set, 1736 .conn_request = tcp_v4_conn_request, 1737 .syn_recv_sock = tcp_v4_syn_recv_sock, 1738 .net_header_len = sizeof(struct iphdr), 1739 .setsockopt = ip_setsockopt, 1740 .getsockopt = ip_getsockopt, 1741 .addr2sockaddr = inet_csk_addr2sockaddr, 1742 .sockaddr_len = sizeof(struct sockaddr_in), 1743 .bind_conflict = inet_csk_bind_conflict, 1744 #ifdef CONFIG_COMPAT 1745 .compat_setsockopt = compat_ip_setsockopt, 1746 .compat_getsockopt = compat_ip_getsockopt, 1747 #endif 1748 .mtu_reduced = tcp_v4_mtu_reduced, 1749 }; 1750 EXPORT_SYMBOL(ipv4_specific); 1751 1752 #ifdef CONFIG_TCP_MD5SIG 1753 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { 1754 .md5_lookup = tcp_v4_md5_lookup, 1755 .calc_md5_hash = tcp_v4_md5_hash_skb, 1756 .md5_parse = tcp_v4_parse_md5_keys, 1757 }; 1758 #endif 1759 1760 /* NOTE: A lot of things set to zero explicitly by call to 1761 * sk_alloc() so need not be done here. 1762 */ 1763 static int tcp_v4_init_sock(struct sock *sk) 1764 { 1765 struct inet_connection_sock *icsk = inet_csk(sk); 1766 1767 tcp_init_sock(sk); 1768 1769 icsk->icsk_af_ops = &ipv4_specific; 1770 1771 #ifdef CONFIG_TCP_MD5SIG 1772 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; 1773 #endif 1774 1775 return 0; 1776 } 1777 1778 void tcp_v4_destroy_sock(struct sock *sk) 1779 { 1780 struct tcp_sock *tp = tcp_sk(sk); 1781 1782 tcp_clear_xmit_timers(sk); 1783 1784 tcp_cleanup_congestion_control(sk); 1785 1786 /* Cleanup up the write buffer. */ 1787 tcp_write_queue_purge(sk); 1788 1789 /* Cleans up our, hopefully empty, out_of_order_queue. */ 1790 __skb_queue_purge(&tp->out_of_order_queue); 1791 1792 #ifdef CONFIG_TCP_MD5SIG 1793 /* Clean up the MD5 key list, if any */ 1794 if (tp->md5sig_info) { 1795 tcp_clear_md5_list(sk); 1796 kfree_rcu(tp->md5sig_info, rcu); 1797 tp->md5sig_info = NULL; 1798 } 1799 #endif 1800 1801 /* Clean prequeue, it must be empty really */ 1802 __skb_queue_purge(&tp->ucopy.prequeue); 1803 1804 /* Clean up a referenced TCP bind bucket. */ 1805 if (inet_csk(sk)->icsk_bind_hash) 1806 inet_put_port(sk); 1807 1808 BUG_ON(tp->fastopen_rsk); 1809 1810 /* If socket is aborted during connect operation */ 1811 tcp_free_fastopen_req(tp); 1812 tcp_saved_syn_free(tp); 1813 1814 sk_sockets_allocated_dec(sk); 1815 sock_release_memcg(sk); 1816 } 1817 EXPORT_SYMBOL(tcp_v4_destroy_sock); 1818 1819 #ifdef CONFIG_PROC_FS 1820 /* Proc filesystem TCP sock list dumping. */ 1821 1822 /* 1823 * Get next listener socket follow cur. If cur is NULL, get first socket 1824 * starting from bucket given in st->bucket; when st->bucket is zero the 1825 * very first socket in the hash table is returned. 1826 */ 1827 static void *listening_get_next(struct seq_file *seq, void *cur) 1828 { 1829 struct inet_connection_sock *icsk; 1830 struct hlist_nulls_node *node; 1831 struct sock *sk = cur; 1832 struct inet_listen_hashbucket *ilb; 1833 struct tcp_iter_state *st = seq->private; 1834 struct net *net = seq_file_net(seq); 1835 1836 if (!sk) { 1837 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1838 spin_lock_bh(&ilb->lock); 1839 sk = sk_nulls_head(&ilb->head); 1840 st->offset = 0; 1841 goto get_sk; 1842 } 1843 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1844 ++st->num; 1845 ++st->offset; 1846 1847 sk = sk_nulls_next(sk); 1848 get_sk: 1849 sk_nulls_for_each_from(sk, node) { 1850 if (!net_eq(sock_net(sk), net)) 1851 continue; 1852 if (sk->sk_family == st->family) { 1853 cur = sk; 1854 goto out; 1855 } 1856 icsk = inet_csk(sk); 1857 } 1858 spin_unlock_bh(&ilb->lock); 1859 st->offset = 0; 1860 if (++st->bucket < INET_LHTABLE_SIZE) { 1861 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1862 spin_lock_bh(&ilb->lock); 1863 sk = sk_nulls_head(&ilb->head); 1864 goto get_sk; 1865 } 1866 cur = NULL; 1867 out: 1868 return cur; 1869 } 1870 1871 static void *listening_get_idx(struct seq_file *seq, loff_t *pos) 1872 { 1873 struct tcp_iter_state *st = seq->private; 1874 void *rc; 1875 1876 st->bucket = 0; 1877 st->offset = 0; 1878 rc = listening_get_next(seq, NULL); 1879 1880 while (rc && *pos) { 1881 rc = listening_get_next(seq, rc); 1882 --*pos; 1883 } 1884 return rc; 1885 } 1886 1887 static inline bool empty_bucket(const struct tcp_iter_state *st) 1888 { 1889 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); 1890 } 1891 1892 /* 1893 * Get first established socket starting from bucket given in st->bucket. 1894 * If st->bucket is zero, the very first socket in the hash is returned. 1895 */ 1896 static void *established_get_first(struct seq_file *seq) 1897 { 1898 struct tcp_iter_state *st = seq->private; 1899 struct net *net = seq_file_net(seq); 1900 void *rc = NULL; 1901 1902 st->offset = 0; 1903 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { 1904 struct sock *sk; 1905 struct hlist_nulls_node *node; 1906 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); 1907 1908 /* Lockless fast path for the common case of empty buckets */ 1909 if (empty_bucket(st)) 1910 continue; 1911 1912 spin_lock_bh(lock); 1913 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { 1914 if (sk->sk_family != st->family || 1915 !net_eq(sock_net(sk), net)) { 1916 continue; 1917 } 1918 rc = sk; 1919 goto out; 1920 } 1921 spin_unlock_bh(lock); 1922 } 1923 out: 1924 return rc; 1925 } 1926 1927 static void *established_get_next(struct seq_file *seq, void *cur) 1928 { 1929 struct sock *sk = cur; 1930 struct hlist_nulls_node *node; 1931 struct tcp_iter_state *st = seq->private; 1932 struct net *net = seq_file_net(seq); 1933 1934 ++st->num; 1935 ++st->offset; 1936 1937 sk = sk_nulls_next(sk); 1938 1939 sk_nulls_for_each_from(sk, node) { 1940 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) 1941 return sk; 1942 } 1943 1944 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); 1945 ++st->bucket; 1946 return established_get_first(seq); 1947 } 1948 1949 static void *established_get_idx(struct seq_file *seq, loff_t pos) 1950 { 1951 struct tcp_iter_state *st = seq->private; 1952 void *rc; 1953 1954 st->bucket = 0; 1955 rc = established_get_first(seq); 1956 1957 while (rc && pos) { 1958 rc = established_get_next(seq, rc); 1959 --pos; 1960 } 1961 return rc; 1962 } 1963 1964 static void *tcp_get_idx(struct seq_file *seq, loff_t pos) 1965 { 1966 void *rc; 1967 struct tcp_iter_state *st = seq->private; 1968 1969 st->state = TCP_SEQ_STATE_LISTENING; 1970 rc = listening_get_idx(seq, &pos); 1971 1972 if (!rc) { 1973 st->state = TCP_SEQ_STATE_ESTABLISHED; 1974 rc = established_get_idx(seq, pos); 1975 } 1976 1977 return rc; 1978 } 1979 1980 static void *tcp_seek_last_pos(struct seq_file *seq) 1981 { 1982 struct tcp_iter_state *st = seq->private; 1983 int offset = st->offset; 1984 int orig_num = st->num; 1985 void *rc = NULL; 1986 1987 switch (st->state) { 1988 case TCP_SEQ_STATE_LISTENING: 1989 if (st->bucket >= INET_LHTABLE_SIZE) 1990 break; 1991 st->state = TCP_SEQ_STATE_LISTENING; 1992 rc = listening_get_next(seq, NULL); 1993 while (offset-- && rc) 1994 rc = listening_get_next(seq, rc); 1995 if (rc) 1996 break; 1997 st->bucket = 0; 1998 st->state = TCP_SEQ_STATE_ESTABLISHED; 1999 /* Fallthrough */ 2000 case TCP_SEQ_STATE_ESTABLISHED: 2001 if (st->bucket > tcp_hashinfo.ehash_mask) 2002 break; 2003 rc = established_get_first(seq); 2004 while (offset-- && rc) 2005 rc = established_get_next(seq, rc); 2006 } 2007 2008 st->num = orig_num; 2009 2010 return rc; 2011 } 2012 2013 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) 2014 { 2015 struct tcp_iter_state *st = seq->private; 2016 void *rc; 2017 2018 if (*pos && *pos == st->last_pos) { 2019 rc = tcp_seek_last_pos(seq); 2020 if (rc) 2021 goto out; 2022 } 2023 2024 st->state = TCP_SEQ_STATE_LISTENING; 2025 st->num = 0; 2026 st->bucket = 0; 2027 st->offset = 0; 2028 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; 2029 2030 out: 2031 st->last_pos = *pos; 2032 return rc; 2033 } 2034 2035 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2036 { 2037 struct tcp_iter_state *st = seq->private; 2038 void *rc = NULL; 2039 2040 if (v == SEQ_START_TOKEN) { 2041 rc = tcp_get_idx(seq, 0); 2042 goto out; 2043 } 2044 2045 switch (st->state) { 2046 case TCP_SEQ_STATE_LISTENING: 2047 rc = listening_get_next(seq, v); 2048 if (!rc) { 2049 st->state = TCP_SEQ_STATE_ESTABLISHED; 2050 st->bucket = 0; 2051 st->offset = 0; 2052 rc = established_get_first(seq); 2053 } 2054 break; 2055 case TCP_SEQ_STATE_ESTABLISHED: 2056 rc = established_get_next(seq, v); 2057 break; 2058 } 2059 out: 2060 ++*pos; 2061 st->last_pos = *pos; 2062 return rc; 2063 } 2064 2065 static void tcp_seq_stop(struct seq_file *seq, void *v) 2066 { 2067 struct tcp_iter_state *st = seq->private; 2068 2069 switch (st->state) { 2070 case TCP_SEQ_STATE_LISTENING: 2071 if (v != SEQ_START_TOKEN) 2072 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); 2073 break; 2074 case TCP_SEQ_STATE_ESTABLISHED: 2075 if (v) 2076 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); 2077 break; 2078 } 2079 } 2080 2081 int tcp_seq_open(struct inode *inode, struct file *file) 2082 { 2083 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode); 2084 struct tcp_iter_state *s; 2085 int err; 2086 2087 err = seq_open_net(inode, file, &afinfo->seq_ops, 2088 sizeof(struct tcp_iter_state)); 2089 if (err < 0) 2090 return err; 2091 2092 s = ((struct seq_file *)file->private_data)->private; 2093 s->family = afinfo->family; 2094 s->last_pos = 0; 2095 return 0; 2096 } 2097 EXPORT_SYMBOL(tcp_seq_open); 2098 2099 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) 2100 { 2101 int rc = 0; 2102 struct proc_dir_entry *p; 2103 2104 afinfo->seq_ops.start = tcp_seq_start; 2105 afinfo->seq_ops.next = tcp_seq_next; 2106 afinfo->seq_ops.stop = tcp_seq_stop; 2107 2108 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2109 afinfo->seq_fops, afinfo); 2110 if (!p) 2111 rc = -ENOMEM; 2112 return rc; 2113 } 2114 EXPORT_SYMBOL(tcp_proc_register); 2115 2116 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) 2117 { 2118 remove_proc_entry(afinfo->name, net->proc_net); 2119 } 2120 EXPORT_SYMBOL(tcp_proc_unregister); 2121 2122 static void get_openreq4(const struct request_sock *req, 2123 struct seq_file *f, int i) 2124 { 2125 const struct inet_request_sock *ireq = inet_rsk(req); 2126 long delta = req->rsk_timer.expires - jiffies; 2127 2128 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 2129 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", 2130 i, 2131 ireq->ir_loc_addr, 2132 ireq->ir_num, 2133 ireq->ir_rmt_addr, 2134 ntohs(ireq->ir_rmt_port), 2135 TCP_SYN_RECV, 2136 0, 0, /* could print option size, but that is af dependent. */ 2137 1, /* timers active (only the expire timer) */ 2138 jiffies_delta_to_clock_t(delta), 2139 req->num_timeout, 2140 from_kuid_munged(seq_user_ns(f), 2141 sock_i_uid(req->rsk_listener)), 2142 0, /* non standard timer */ 2143 0, /* open_requests have no inode */ 2144 0, 2145 req); 2146 } 2147 2148 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) 2149 { 2150 int timer_active; 2151 unsigned long timer_expires; 2152 const struct tcp_sock *tp = tcp_sk(sk); 2153 const struct inet_connection_sock *icsk = inet_csk(sk); 2154 const struct inet_sock *inet = inet_sk(sk); 2155 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; 2156 __be32 dest = inet->inet_daddr; 2157 __be32 src = inet->inet_rcv_saddr; 2158 __u16 destp = ntohs(inet->inet_dport); 2159 __u16 srcp = ntohs(inet->inet_sport); 2160 int rx_queue; 2161 int state; 2162 2163 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 2164 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || 2165 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 2166 timer_active = 1; 2167 timer_expires = icsk->icsk_timeout; 2168 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 2169 timer_active = 4; 2170 timer_expires = icsk->icsk_timeout; 2171 } else if (timer_pending(&sk->sk_timer)) { 2172 timer_active = 2; 2173 timer_expires = sk->sk_timer.expires; 2174 } else { 2175 timer_active = 0; 2176 timer_expires = jiffies; 2177 } 2178 2179 state = sk_state_load(sk); 2180 if (state == TCP_LISTEN) 2181 rx_queue = sk->sk_ack_backlog; 2182 else 2183 /* Because we don't lock the socket, 2184 * we might find a transient negative value. 2185 */ 2186 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); 2187 2188 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " 2189 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", 2190 i, src, srcp, dest, destp, state, 2191 tp->write_seq - tp->snd_una, 2192 rx_queue, 2193 timer_active, 2194 jiffies_delta_to_clock_t(timer_expires - jiffies), 2195 icsk->icsk_retransmits, 2196 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), 2197 icsk->icsk_probes_out, 2198 sock_i_ino(sk), 2199 atomic_read(&sk->sk_refcnt), sk, 2200 jiffies_to_clock_t(icsk->icsk_rto), 2201 jiffies_to_clock_t(icsk->icsk_ack.ato), 2202 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, 2203 tp->snd_cwnd, 2204 state == TCP_LISTEN ? 2205 fastopenq->max_qlen : 2206 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); 2207 } 2208 2209 static void get_timewait4_sock(const struct inet_timewait_sock *tw, 2210 struct seq_file *f, int i) 2211 { 2212 long delta = tw->tw_timer.expires - jiffies; 2213 __be32 dest, src; 2214 __u16 destp, srcp; 2215 2216 dest = tw->tw_daddr; 2217 src = tw->tw_rcv_saddr; 2218 destp = ntohs(tw->tw_dport); 2219 srcp = ntohs(tw->tw_sport); 2220 2221 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 2222 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", 2223 i, src, srcp, dest, destp, tw->tw_substate, 0, 0, 2224 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2225 atomic_read(&tw->tw_refcnt), tw); 2226 } 2227 2228 #define TMPSZ 150 2229 2230 static int tcp4_seq_show(struct seq_file *seq, void *v) 2231 { 2232 struct tcp_iter_state *st; 2233 struct sock *sk = v; 2234 2235 seq_setwidth(seq, TMPSZ - 1); 2236 if (v == SEQ_START_TOKEN) { 2237 seq_puts(seq, " sl local_address rem_address st tx_queue " 2238 "rx_queue tr tm->when retrnsmt uid timeout " 2239 "inode"); 2240 goto out; 2241 } 2242 st = seq->private; 2243 2244 if (sk->sk_state == TCP_TIME_WAIT) 2245 get_timewait4_sock(v, seq, st->num); 2246 else if (sk->sk_state == TCP_NEW_SYN_RECV) 2247 get_openreq4(v, seq, st->num); 2248 else 2249 get_tcp4_sock(v, seq, st->num); 2250 out: 2251 seq_pad(seq, '\n'); 2252 return 0; 2253 } 2254 2255 static const struct file_operations tcp_afinfo_seq_fops = { 2256 .owner = THIS_MODULE, 2257 .open = tcp_seq_open, 2258 .read = seq_read, 2259 .llseek = seq_lseek, 2260 .release = seq_release_net 2261 }; 2262 2263 static struct tcp_seq_afinfo tcp4_seq_afinfo = { 2264 .name = "tcp", 2265 .family = AF_INET, 2266 .seq_fops = &tcp_afinfo_seq_fops, 2267 .seq_ops = { 2268 .show = tcp4_seq_show, 2269 }, 2270 }; 2271 2272 static int __net_init tcp4_proc_init_net(struct net *net) 2273 { 2274 return tcp_proc_register(net, &tcp4_seq_afinfo); 2275 } 2276 2277 static void __net_exit tcp4_proc_exit_net(struct net *net) 2278 { 2279 tcp_proc_unregister(net, &tcp4_seq_afinfo); 2280 } 2281 2282 static struct pernet_operations tcp4_net_ops = { 2283 .init = tcp4_proc_init_net, 2284 .exit = tcp4_proc_exit_net, 2285 }; 2286 2287 int __init tcp4_proc_init(void) 2288 { 2289 return register_pernet_subsys(&tcp4_net_ops); 2290 } 2291 2292 void tcp4_proc_exit(void) 2293 { 2294 unregister_pernet_subsys(&tcp4_net_ops); 2295 } 2296 #endif /* CONFIG_PROC_FS */ 2297 2298 struct proto tcp_prot = { 2299 .name = "TCP", 2300 .owner = THIS_MODULE, 2301 .close = tcp_close, 2302 .connect = tcp_v4_connect, 2303 .disconnect = tcp_disconnect, 2304 .accept = inet_csk_accept, 2305 .ioctl = tcp_ioctl, 2306 .init = tcp_v4_init_sock, 2307 .destroy = tcp_v4_destroy_sock, 2308 .shutdown = tcp_shutdown, 2309 .setsockopt = tcp_setsockopt, 2310 .getsockopt = tcp_getsockopt, 2311 .recvmsg = tcp_recvmsg, 2312 .sendmsg = tcp_sendmsg, 2313 .sendpage = tcp_sendpage, 2314 .backlog_rcv = tcp_v4_do_rcv, 2315 .release_cb = tcp_release_cb, 2316 .hash = inet_hash, 2317 .unhash = inet_unhash, 2318 .get_port = inet_csk_get_port, 2319 .enter_memory_pressure = tcp_enter_memory_pressure, 2320 .stream_memory_free = tcp_stream_memory_free, 2321 .sockets_allocated = &tcp_sockets_allocated, 2322 .orphan_count = &tcp_orphan_count, 2323 .memory_allocated = &tcp_memory_allocated, 2324 .memory_pressure = &tcp_memory_pressure, 2325 .sysctl_mem = sysctl_tcp_mem, 2326 .sysctl_wmem = sysctl_tcp_wmem, 2327 .sysctl_rmem = sysctl_tcp_rmem, 2328 .max_header = MAX_TCP_HEADER, 2329 .obj_size = sizeof(struct tcp_sock), 2330 .slab_flags = SLAB_DESTROY_BY_RCU, 2331 .twsk_prot = &tcp_timewait_sock_ops, 2332 .rsk_prot = &tcp_request_sock_ops, 2333 .h.hashinfo = &tcp_hashinfo, 2334 .no_autobind = true, 2335 #ifdef CONFIG_COMPAT 2336 .compat_setsockopt = compat_tcp_setsockopt, 2337 .compat_getsockopt = compat_tcp_getsockopt, 2338 #endif 2339 #ifdef CONFIG_MEMCG_KMEM 2340 .init_cgroup = tcp_init_cgroup, 2341 .destroy_cgroup = tcp_destroy_cgroup, 2342 .proto_cgroup = tcp_proto_cgroup, 2343 #endif 2344 }; 2345 EXPORT_SYMBOL(tcp_prot); 2346 2347 static void __net_exit tcp_sk_exit(struct net *net) 2348 { 2349 int cpu; 2350 2351 for_each_possible_cpu(cpu) 2352 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu)); 2353 free_percpu(net->ipv4.tcp_sk); 2354 } 2355 2356 static int __net_init tcp_sk_init(struct net *net) 2357 { 2358 int res, cpu; 2359 2360 net->ipv4.tcp_sk = alloc_percpu(struct sock *); 2361 if (!net->ipv4.tcp_sk) 2362 return -ENOMEM; 2363 2364 for_each_possible_cpu(cpu) { 2365 struct sock *sk; 2366 2367 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW, 2368 IPPROTO_TCP, net); 2369 if (res) 2370 goto fail; 2371 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk; 2372 } 2373 2374 net->ipv4.sysctl_tcp_ecn = 2; 2375 net->ipv4.sysctl_tcp_ecn_fallback = 1; 2376 2377 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; 2378 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD; 2379 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL; 2380 2381 return 0; 2382 fail: 2383 tcp_sk_exit(net); 2384 2385 return res; 2386 } 2387 2388 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) 2389 { 2390 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); 2391 } 2392 2393 static struct pernet_operations __net_initdata tcp_sk_ops = { 2394 .init = tcp_sk_init, 2395 .exit = tcp_sk_exit, 2396 .exit_batch = tcp_sk_exit_batch, 2397 }; 2398 2399 void __init tcp_v4_init(void) 2400 { 2401 inet_hashinfo_init(&tcp_hashinfo); 2402 if (register_pernet_subsys(&tcp_sk_ops)) 2403 panic("Failed to create the TCP control socket.\n"); 2404 } 2405