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 inet_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 reqsk_put(req); 328 } else { 329 /* 330 * Still in SYN_RECV, just remove it silently. 331 * There is no good way to pass the error to the newly 332 * created socket, and POSIX does not want network 333 * errors returned from accept(). 334 */ 335 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS); 336 inet_csk_reqsk_queue_drop(req->rsk_listener, req); 337 } 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(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(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->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(struct sock *sk, struct dst_entry *dst, 822 struct flowi *fl, 823 struct request_sock *req, 824 u16 queue_mapping, 825 struct tcp_fastopen_cookie *foc) 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); 837 838 if (skb) { 839 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); 840 841 skb_set_queue_mapping(skb, queue_mapping); 842 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, 843 ireq->ir_rmt_addr, 844 ireq->opt); 845 err = net_xmit_eval(err); 846 } 847 848 return err; 849 } 850 851 /* 852 * IPv4 request_sock destructor. 853 */ 854 static void tcp_v4_reqsk_destructor(struct request_sock *req) 855 { 856 kfree(inet_rsk(req)->opt); 857 } 858 859 860 #ifdef CONFIG_TCP_MD5SIG 861 /* 862 * RFC2385 MD5 checksumming requires a mapping of 863 * IP address->MD5 Key. 864 * We need to maintain these in the sk structure. 865 */ 866 867 /* Find the Key structure for an address. */ 868 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 869 const union tcp_md5_addr *addr, 870 int family) 871 { 872 const struct tcp_sock *tp = tcp_sk(sk); 873 struct tcp_md5sig_key *key; 874 unsigned int size = sizeof(struct in_addr); 875 const struct tcp_md5sig_info *md5sig; 876 877 /* caller either holds rcu_read_lock() or socket lock */ 878 md5sig = rcu_dereference_check(tp->md5sig_info, 879 sock_owned_by_user(sk) || 880 lockdep_is_held(&sk->sk_lock.slock)); 881 if (!md5sig) 882 return NULL; 883 #if IS_ENABLED(CONFIG_IPV6) 884 if (family == AF_INET6) 885 size = sizeof(struct in6_addr); 886 #endif 887 hlist_for_each_entry_rcu(key, &md5sig->head, node) { 888 if (key->family != family) 889 continue; 890 if (!memcmp(&key->addr, addr, size)) 891 return key; 892 } 893 return NULL; 894 } 895 EXPORT_SYMBOL(tcp_md5_do_lookup); 896 897 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 898 const struct sock *addr_sk) 899 { 900 const union tcp_md5_addr *addr; 901 902 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr; 903 return tcp_md5_do_lookup(sk, addr, AF_INET); 904 } 905 EXPORT_SYMBOL(tcp_v4_md5_lookup); 906 907 /* This can be called on a newly created socket, from other files */ 908 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 909 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) 910 { 911 /* Add Key to the list */ 912 struct tcp_md5sig_key *key; 913 struct tcp_sock *tp = tcp_sk(sk); 914 struct tcp_md5sig_info *md5sig; 915 916 key = tcp_md5_do_lookup(sk, addr, family); 917 if (key) { 918 /* Pre-existing entry - just update that one. */ 919 memcpy(key->key, newkey, newkeylen); 920 key->keylen = newkeylen; 921 return 0; 922 } 923 924 md5sig = rcu_dereference_protected(tp->md5sig_info, 925 sock_owned_by_user(sk)); 926 if (!md5sig) { 927 md5sig = kmalloc(sizeof(*md5sig), gfp); 928 if (!md5sig) 929 return -ENOMEM; 930 931 sk_nocaps_add(sk, NETIF_F_GSO_MASK); 932 INIT_HLIST_HEAD(&md5sig->head); 933 rcu_assign_pointer(tp->md5sig_info, md5sig); 934 } 935 936 key = sock_kmalloc(sk, sizeof(*key), gfp); 937 if (!key) 938 return -ENOMEM; 939 if (!tcp_alloc_md5sig_pool()) { 940 sock_kfree_s(sk, key, sizeof(*key)); 941 return -ENOMEM; 942 } 943 944 memcpy(key->key, newkey, newkeylen); 945 key->keylen = newkeylen; 946 key->family = family; 947 memcpy(&key->addr, addr, 948 (family == AF_INET6) ? sizeof(struct in6_addr) : 949 sizeof(struct in_addr)); 950 hlist_add_head_rcu(&key->node, &md5sig->head); 951 return 0; 952 } 953 EXPORT_SYMBOL(tcp_md5_do_add); 954 955 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) 956 { 957 struct tcp_md5sig_key *key; 958 959 key = tcp_md5_do_lookup(sk, addr, family); 960 if (!key) 961 return -ENOENT; 962 hlist_del_rcu(&key->node); 963 atomic_sub(sizeof(*key), &sk->sk_omem_alloc); 964 kfree_rcu(key, rcu); 965 return 0; 966 } 967 EXPORT_SYMBOL(tcp_md5_do_del); 968 969 static void tcp_clear_md5_list(struct sock *sk) 970 { 971 struct tcp_sock *tp = tcp_sk(sk); 972 struct tcp_md5sig_key *key; 973 struct hlist_node *n; 974 struct tcp_md5sig_info *md5sig; 975 976 md5sig = rcu_dereference_protected(tp->md5sig_info, 1); 977 978 hlist_for_each_entry_safe(key, n, &md5sig->head, node) { 979 hlist_del_rcu(&key->node); 980 atomic_sub(sizeof(*key), &sk->sk_omem_alloc); 981 kfree_rcu(key, rcu); 982 } 983 } 984 985 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, 986 int optlen) 987 { 988 struct tcp_md5sig cmd; 989 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; 990 991 if (optlen < sizeof(cmd)) 992 return -EINVAL; 993 994 if (copy_from_user(&cmd, optval, sizeof(cmd))) 995 return -EFAULT; 996 997 if (sin->sin_family != AF_INET) 998 return -EINVAL; 999 1000 if (!cmd.tcpm_keylen) 1001 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, 1002 AF_INET); 1003 1004 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) 1005 return -EINVAL; 1006 1007 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, 1008 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, 1009 GFP_KERNEL); 1010 } 1011 1012 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, 1013 __be32 daddr, __be32 saddr, int nbytes) 1014 { 1015 struct tcp4_pseudohdr *bp; 1016 struct scatterlist sg; 1017 1018 bp = &hp->md5_blk.ip4; 1019 1020 /* 1021 * 1. the TCP pseudo-header (in the order: source IP address, 1022 * destination IP address, zero-padded protocol number, and 1023 * segment length) 1024 */ 1025 bp->saddr = saddr; 1026 bp->daddr = daddr; 1027 bp->pad = 0; 1028 bp->protocol = IPPROTO_TCP; 1029 bp->len = cpu_to_be16(nbytes); 1030 1031 sg_init_one(&sg, bp, sizeof(*bp)); 1032 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); 1033 } 1034 1035 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, 1036 __be32 daddr, __be32 saddr, const struct tcphdr *th) 1037 { 1038 struct tcp_md5sig_pool *hp; 1039 struct hash_desc *desc; 1040 1041 hp = tcp_get_md5sig_pool(); 1042 if (!hp) 1043 goto clear_hash_noput; 1044 desc = &hp->md5_desc; 1045 1046 if (crypto_hash_init(desc)) 1047 goto clear_hash; 1048 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) 1049 goto clear_hash; 1050 if (tcp_md5_hash_header(hp, th)) 1051 goto clear_hash; 1052 if (tcp_md5_hash_key(hp, key)) 1053 goto clear_hash; 1054 if (crypto_hash_final(desc, md5_hash)) 1055 goto clear_hash; 1056 1057 tcp_put_md5sig_pool(); 1058 return 0; 1059 1060 clear_hash: 1061 tcp_put_md5sig_pool(); 1062 clear_hash_noput: 1063 memset(md5_hash, 0, 16); 1064 return 1; 1065 } 1066 1067 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1068 const struct sock *sk, 1069 const struct sk_buff *skb) 1070 { 1071 struct tcp_md5sig_pool *hp; 1072 struct hash_desc *desc; 1073 const struct tcphdr *th = tcp_hdr(skb); 1074 __be32 saddr, daddr; 1075 1076 if (sk) { /* valid for establish/request sockets */ 1077 saddr = sk->sk_rcv_saddr; 1078 daddr = sk->sk_daddr; 1079 } else { 1080 const struct iphdr *iph = ip_hdr(skb); 1081 saddr = iph->saddr; 1082 daddr = iph->daddr; 1083 } 1084 1085 hp = tcp_get_md5sig_pool(); 1086 if (!hp) 1087 goto clear_hash_noput; 1088 desc = &hp->md5_desc; 1089 1090 if (crypto_hash_init(desc)) 1091 goto clear_hash; 1092 1093 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) 1094 goto clear_hash; 1095 if (tcp_md5_hash_header(hp, th)) 1096 goto clear_hash; 1097 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) 1098 goto clear_hash; 1099 if (tcp_md5_hash_key(hp, key)) 1100 goto clear_hash; 1101 if (crypto_hash_final(desc, md5_hash)) 1102 goto clear_hash; 1103 1104 tcp_put_md5sig_pool(); 1105 return 0; 1106 1107 clear_hash: 1108 tcp_put_md5sig_pool(); 1109 clear_hash_noput: 1110 memset(md5_hash, 0, 16); 1111 return 1; 1112 } 1113 EXPORT_SYMBOL(tcp_v4_md5_hash_skb); 1114 1115 /* Called with rcu_read_lock() */ 1116 static bool tcp_v4_inbound_md5_hash(struct sock *sk, 1117 const struct sk_buff *skb) 1118 { 1119 /* 1120 * This gets called for each TCP segment that arrives 1121 * so we want to be efficient. 1122 * We have 3 drop cases: 1123 * o No MD5 hash and one expected. 1124 * o MD5 hash and we're not expecting one. 1125 * o MD5 hash and its wrong. 1126 */ 1127 const __u8 *hash_location = NULL; 1128 struct tcp_md5sig_key *hash_expected; 1129 const struct iphdr *iph = ip_hdr(skb); 1130 const struct tcphdr *th = tcp_hdr(skb); 1131 int genhash; 1132 unsigned char newhash[16]; 1133 1134 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, 1135 AF_INET); 1136 hash_location = tcp_parse_md5sig_option(th); 1137 1138 /* We've parsed the options - do we have a hash? */ 1139 if (!hash_expected && !hash_location) 1140 return false; 1141 1142 if (hash_expected && !hash_location) { 1143 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 1144 return true; 1145 } 1146 1147 if (!hash_expected && hash_location) { 1148 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); 1149 return true; 1150 } 1151 1152 /* Okay, so this is hash_expected and hash_location - 1153 * so we need to calculate the checksum. 1154 */ 1155 genhash = tcp_v4_md5_hash_skb(newhash, 1156 hash_expected, 1157 NULL, skb); 1158 1159 if (genhash || memcmp(hash_location, newhash, 16) != 0) { 1160 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", 1161 &iph->saddr, ntohs(th->source), 1162 &iph->daddr, ntohs(th->dest), 1163 genhash ? " tcp_v4_calc_md5_hash failed" 1164 : ""); 1165 return true; 1166 } 1167 return false; 1168 } 1169 #endif 1170 1171 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk_listener, 1172 struct sk_buff *skb) 1173 { 1174 struct inet_request_sock *ireq = inet_rsk(req); 1175 1176 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); 1177 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); 1178 ireq->no_srccheck = inet_sk(sk_listener)->transparent; 1179 ireq->opt = tcp_v4_save_options(skb); 1180 } 1181 1182 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl, 1183 const struct request_sock *req, 1184 bool *strict) 1185 { 1186 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req); 1187 1188 if (strict) { 1189 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr) 1190 *strict = true; 1191 else 1192 *strict = false; 1193 } 1194 1195 return dst; 1196 } 1197 1198 struct request_sock_ops tcp_request_sock_ops __read_mostly = { 1199 .family = PF_INET, 1200 .obj_size = sizeof(struct tcp_request_sock), 1201 .rtx_syn_ack = tcp_rtx_synack, 1202 .send_ack = tcp_v4_reqsk_send_ack, 1203 .destructor = tcp_v4_reqsk_destructor, 1204 .send_reset = tcp_v4_send_reset, 1205 .syn_ack_timeout = tcp_syn_ack_timeout, 1206 }; 1207 1208 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { 1209 .mss_clamp = TCP_MSS_DEFAULT, 1210 #ifdef CONFIG_TCP_MD5SIG 1211 .req_md5_lookup = tcp_v4_md5_lookup, 1212 .calc_md5_hash = tcp_v4_md5_hash_skb, 1213 #endif 1214 .init_req = tcp_v4_init_req, 1215 #ifdef CONFIG_SYN_COOKIES 1216 .cookie_init_seq = cookie_v4_init_sequence, 1217 #endif 1218 .route_req = tcp_v4_route_req, 1219 .init_seq = tcp_v4_init_sequence, 1220 .send_synack = tcp_v4_send_synack, 1221 .queue_hash_add = inet_csk_reqsk_queue_hash_add, 1222 }; 1223 1224 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) 1225 { 1226 /* Never answer to SYNs send to broadcast or multicast */ 1227 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 1228 goto drop; 1229 1230 return tcp_conn_request(&tcp_request_sock_ops, 1231 &tcp_request_sock_ipv4_ops, sk, skb); 1232 1233 drop: 1234 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 1235 return 0; 1236 } 1237 EXPORT_SYMBOL(tcp_v4_conn_request); 1238 1239 1240 /* 1241 * The three way handshake has completed - we got a valid synack - 1242 * now create the new socket. 1243 */ 1244 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 1245 struct request_sock *req, 1246 struct dst_entry *dst) 1247 { 1248 struct inet_request_sock *ireq; 1249 struct inet_sock *newinet; 1250 struct tcp_sock *newtp; 1251 struct sock *newsk; 1252 #ifdef CONFIG_TCP_MD5SIG 1253 struct tcp_md5sig_key *key; 1254 #endif 1255 struct ip_options_rcu *inet_opt; 1256 1257 if (sk_acceptq_is_full(sk)) 1258 goto exit_overflow; 1259 1260 newsk = tcp_create_openreq_child(sk, req, skb); 1261 if (!newsk) 1262 goto exit_nonewsk; 1263 1264 newsk->sk_gso_type = SKB_GSO_TCPV4; 1265 inet_sk_rx_dst_set(newsk, skb); 1266 1267 newtp = tcp_sk(newsk); 1268 newinet = inet_sk(newsk); 1269 ireq = inet_rsk(req); 1270 sk_daddr_set(newsk, ireq->ir_rmt_addr); 1271 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); 1272 newinet->inet_saddr = ireq->ir_loc_addr; 1273 inet_opt = ireq->opt; 1274 rcu_assign_pointer(newinet->inet_opt, inet_opt); 1275 ireq->opt = NULL; 1276 newinet->mc_index = inet_iif(skb); 1277 newinet->mc_ttl = ip_hdr(skb)->ttl; 1278 newinet->rcv_tos = ip_hdr(skb)->tos; 1279 inet_csk(newsk)->icsk_ext_hdr_len = 0; 1280 inet_set_txhash(newsk); 1281 if (inet_opt) 1282 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; 1283 newinet->inet_id = newtp->write_seq ^ jiffies; 1284 1285 if (!dst) { 1286 dst = inet_csk_route_child_sock(sk, newsk, req); 1287 if (!dst) 1288 goto put_and_exit; 1289 } else { 1290 /* syncookie case : see end of cookie_v4_check() */ 1291 } 1292 sk_setup_caps(newsk, dst); 1293 1294 tcp_ca_openreq_child(newsk, dst); 1295 1296 tcp_sync_mss(newsk, dst_mtu(dst)); 1297 newtp->advmss = dst_metric_advmss(dst); 1298 if (tcp_sk(sk)->rx_opt.user_mss && 1299 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) 1300 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; 1301 1302 tcp_initialize_rcv_mss(newsk); 1303 1304 #ifdef CONFIG_TCP_MD5SIG 1305 /* Copy over the MD5 key from the original socket */ 1306 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, 1307 AF_INET); 1308 if (key) { 1309 /* 1310 * We're using one, so create a matching key 1311 * on the newsk structure. If we fail to get 1312 * memory, then we end up not copying the key 1313 * across. Shucks. 1314 */ 1315 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, 1316 AF_INET, key->key, key->keylen, GFP_ATOMIC); 1317 sk_nocaps_add(newsk, NETIF_F_GSO_MASK); 1318 } 1319 #endif 1320 1321 if (__inet_inherit_port(sk, newsk) < 0) 1322 goto put_and_exit; 1323 __inet_hash_nolisten(newsk, NULL); 1324 1325 return newsk; 1326 1327 exit_overflow: 1328 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 1329 exit_nonewsk: 1330 dst_release(dst); 1331 exit: 1332 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 1333 return NULL; 1334 put_and_exit: 1335 inet_csk_prepare_forced_close(newsk); 1336 tcp_done(newsk); 1337 goto exit; 1338 } 1339 EXPORT_SYMBOL(tcp_v4_syn_recv_sock); 1340 1341 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) 1342 { 1343 const struct tcphdr *th = tcp_hdr(skb); 1344 const struct iphdr *iph = ip_hdr(skb); 1345 struct request_sock *req; 1346 struct sock *nsk; 1347 1348 req = inet_csk_search_req(sk, th->source, iph->saddr, iph->daddr); 1349 if (req) { 1350 nsk = tcp_check_req(sk, skb, req, false); 1351 if (!nsk) 1352 reqsk_put(req); 1353 return nsk; 1354 } 1355 1356 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, 1357 th->source, iph->daddr, th->dest, inet_iif(skb)); 1358 1359 if (nsk) { 1360 if (nsk->sk_state != TCP_TIME_WAIT) { 1361 bh_lock_sock(nsk); 1362 return nsk; 1363 } 1364 inet_twsk_put(inet_twsk(nsk)); 1365 return NULL; 1366 } 1367 1368 #ifdef CONFIG_SYN_COOKIES 1369 if (!th->syn) 1370 sk = cookie_v4_check(sk, skb); 1371 #endif 1372 return sk; 1373 } 1374 1375 /* The socket must have it's spinlock held when we get 1376 * here. 1377 * 1378 * We have a potential double-lock case here, so even when 1379 * doing backlog processing we use the BH locking scheme. 1380 * This is because we cannot sleep with the original spinlock 1381 * held. 1382 */ 1383 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) 1384 { 1385 struct sock *rsk; 1386 1387 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ 1388 struct dst_entry *dst = sk->sk_rx_dst; 1389 1390 sock_rps_save_rxhash(sk, skb); 1391 sk_mark_napi_id(sk, skb); 1392 if (dst) { 1393 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || 1394 !dst->ops->check(dst, 0)) { 1395 dst_release(dst); 1396 sk->sk_rx_dst = NULL; 1397 } 1398 } 1399 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); 1400 return 0; 1401 } 1402 1403 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) 1404 goto csum_err; 1405 1406 if (sk->sk_state == TCP_LISTEN) { 1407 struct sock *nsk = tcp_v4_hnd_req(sk, skb); 1408 if (!nsk) 1409 goto discard; 1410 1411 if (nsk != sk) { 1412 sock_rps_save_rxhash(nsk, skb); 1413 sk_mark_napi_id(sk, skb); 1414 if (tcp_child_process(sk, nsk, skb)) { 1415 rsk = nsk; 1416 goto reset; 1417 } 1418 return 0; 1419 } 1420 } else 1421 sock_rps_save_rxhash(sk, skb); 1422 1423 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { 1424 rsk = sk; 1425 goto reset; 1426 } 1427 return 0; 1428 1429 reset: 1430 tcp_v4_send_reset(rsk, skb); 1431 discard: 1432 kfree_skb(skb); 1433 /* Be careful here. If this function gets more complicated and 1434 * gcc suffers from register pressure on the x86, sk (in %ebx) 1435 * might be destroyed here. This current version compiles correctly, 1436 * but you have been warned. 1437 */ 1438 return 0; 1439 1440 csum_err: 1441 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); 1442 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); 1443 goto discard; 1444 } 1445 EXPORT_SYMBOL(tcp_v4_do_rcv); 1446 1447 void tcp_v4_early_demux(struct sk_buff *skb) 1448 { 1449 const struct iphdr *iph; 1450 const struct tcphdr *th; 1451 struct sock *sk; 1452 1453 if (skb->pkt_type != PACKET_HOST) 1454 return; 1455 1456 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) 1457 return; 1458 1459 iph = ip_hdr(skb); 1460 th = tcp_hdr(skb); 1461 1462 if (th->doff < sizeof(struct tcphdr) / 4) 1463 return; 1464 1465 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, 1466 iph->saddr, th->source, 1467 iph->daddr, ntohs(th->dest), 1468 skb->skb_iif); 1469 if (sk) { 1470 skb->sk = sk; 1471 skb->destructor = sock_edemux; 1472 if (sk_fullsock(sk)) { 1473 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); 1474 1475 if (dst) 1476 dst = dst_check(dst, 0); 1477 if (dst && 1478 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) 1479 skb_dst_set_noref(skb, dst); 1480 } 1481 } 1482 } 1483 1484 /* Packet is added to VJ-style prequeue for processing in process 1485 * context, if a reader task is waiting. Apparently, this exciting 1486 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 1487 * failed somewhere. Latency? Burstiness? Well, at least now we will 1488 * see, why it failed. 8)8) --ANK 1489 * 1490 */ 1491 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) 1492 { 1493 struct tcp_sock *tp = tcp_sk(sk); 1494 1495 if (sysctl_tcp_low_latency || !tp->ucopy.task) 1496 return false; 1497 1498 if (skb->len <= tcp_hdrlen(skb) && 1499 skb_queue_len(&tp->ucopy.prequeue) == 0) 1500 return false; 1501 1502 /* Before escaping RCU protected region, we need to take care of skb 1503 * dst. Prequeue is only enabled for established sockets. 1504 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst 1505 * Instead of doing full sk_rx_dst validity here, let's perform 1506 * an optimistic check. 1507 */ 1508 if (likely(sk->sk_rx_dst)) 1509 skb_dst_drop(skb); 1510 else 1511 skb_dst_force(skb); 1512 1513 __skb_queue_tail(&tp->ucopy.prequeue, skb); 1514 tp->ucopy.memory += skb->truesize; 1515 if (tp->ucopy.memory > sk->sk_rcvbuf) { 1516 struct sk_buff *skb1; 1517 1518 BUG_ON(sock_owned_by_user(sk)); 1519 1520 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 1521 sk_backlog_rcv(sk, skb1); 1522 NET_INC_STATS_BH(sock_net(sk), 1523 LINUX_MIB_TCPPREQUEUEDROPPED); 1524 } 1525 1526 tp->ucopy.memory = 0; 1527 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 1528 wake_up_interruptible_sync_poll(sk_sleep(sk), 1529 POLLIN | POLLRDNORM | POLLRDBAND); 1530 if (!inet_csk_ack_scheduled(sk)) 1531 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 1532 (3 * tcp_rto_min(sk)) / 4, 1533 TCP_RTO_MAX); 1534 } 1535 return true; 1536 } 1537 EXPORT_SYMBOL(tcp_prequeue); 1538 1539 /* 1540 * From tcp_input.c 1541 */ 1542 1543 int tcp_v4_rcv(struct sk_buff *skb) 1544 { 1545 const struct iphdr *iph; 1546 const struct tcphdr *th; 1547 struct sock *sk; 1548 int ret; 1549 struct net *net = dev_net(skb->dev); 1550 1551 if (skb->pkt_type != PACKET_HOST) 1552 goto discard_it; 1553 1554 /* Count it even if it's bad */ 1555 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); 1556 1557 if (!pskb_may_pull(skb, sizeof(struct tcphdr))) 1558 goto discard_it; 1559 1560 th = tcp_hdr(skb); 1561 1562 if (th->doff < sizeof(struct tcphdr) / 4) 1563 goto bad_packet; 1564 if (!pskb_may_pull(skb, th->doff * 4)) 1565 goto discard_it; 1566 1567 /* An explanation is required here, I think. 1568 * Packet length and doff are validated by header prediction, 1569 * provided case of th->doff==0 is eliminated. 1570 * So, we defer the checks. */ 1571 1572 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) 1573 goto csum_error; 1574 1575 th = tcp_hdr(skb); 1576 iph = ip_hdr(skb); 1577 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB() 1578 * barrier() makes sure compiler wont play fool^Waliasing games. 1579 */ 1580 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb), 1581 sizeof(struct inet_skb_parm)); 1582 barrier(); 1583 1584 TCP_SKB_CB(skb)->seq = ntohl(th->seq); 1585 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + 1586 skb->len - th->doff * 4); 1587 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); 1588 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); 1589 TCP_SKB_CB(skb)->tcp_tw_isn = 0; 1590 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); 1591 TCP_SKB_CB(skb)->sacked = 0; 1592 1593 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); 1594 if (!sk) 1595 goto no_tcp_socket; 1596 1597 process: 1598 if (sk->sk_state == TCP_TIME_WAIT) 1599 goto do_time_wait; 1600 1601 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { 1602 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); 1603 goto discard_and_relse; 1604 } 1605 1606 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1607 goto discard_and_relse; 1608 1609 #ifdef CONFIG_TCP_MD5SIG 1610 /* 1611 * We really want to reject the packet as early as possible 1612 * if: 1613 * o We're expecting an MD5'd packet and this is no MD5 tcp option 1614 * o There is an MD5 option and we're not expecting one 1615 */ 1616 if (tcp_v4_inbound_md5_hash(sk, skb)) 1617 goto discard_and_relse; 1618 #endif 1619 1620 nf_reset(skb); 1621 1622 if (sk_filter(sk, skb)) 1623 goto discard_and_relse; 1624 1625 sk_incoming_cpu_update(sk); 1626 skb->dev = NULL; 1627 1628 bh_lock_sock_nested(sk); 1629 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1630 ret = 0; 1631 if (!sock_owned_by_user(sk)) { 1632 if (!tcp_prequeue(sk, skb)) 1633 ret = tcp_v4_do_rcv(sk, skb); 1634 } else if (unlikely(sk_add_backlog(sk, skb, 1635 sk->sk_rcvbuf + sk->sk_sndbuf))) { 1636 bh_unlock_sock(sk); 1637 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); 1638 goto discard_and_relse; 1639 } 1640 bh_unlock_sock(sk); 1641 1642 sock_put(sk); 1643 1644 return ret; 1645 1646 no_tcp_socket: 1647 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1648 goto discard_it; 1649 1650 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { 1651 csum_error: 1652 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS); 1653 bad_packet: 1654 TCP_INC_STATS_BH(net, TCP_MIB_INERRS); 1655 } else { 1656 tcp_v4_send_reset(NULL, skb); 1657 } 1658 1659 discard_it: 1660 /* Discard frame. */ 1661 kfree_skb(skb); 1662 return 0; 1663 1664 discard_and_relse: 1665 sock_put(sk); 1666 goto discard_it; 1667 1668 do_time_wait: 1669 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1670 inet_twsk_put(inet_twsk(sk)); 1671 goto discard_it; 1672 } 1673 1674 if (skb->len < (th->doff << 2)) { 1675 inet_twsk_put(inet_twsk(sk)); 1676 goto bad_packet; 1677 } 1678 if (tcp_checksum_complete(skb)) { 1679 inet_twsk_put(inet_twsk(sk)); 1680 goto csum_error; 1681 } 1682 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { 1683 case TCP_TW_SYN: { 1684 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), 1685 &tcp_hashinfo, 1686 iph->saddr, th->source, 1687 iph->daddr, th->dest, 1688 inet_iif(skb)); 1689 if (sk2) { 1690 inet_twsk_deschedule(inet_twsk(sk)); 1691 inet_twsk_put(inet_twsk(sk)); 1692 sk = sk2; 1693 goto process; 1694 } 1695 /* Fall through to ACK */ 1696 } 1697 case TCP_TW_ACK: 1698 tcp_v4_timewait_ack(sk, skb); 1699 break; 1700 case TCP_TW_RST: 1701 goto no_tcp_socket; 1702 case TCP_TW_SUCCESS:; 1703 } 1704 goto discard_it; 1705 } 1706 1707 static struct timewait_sock_ops tcp_timewait_sock_ops = { 1708 .twsk_obj_size = sizeof(struct tcp_timewait_sock), 1709 .twsk_unique = tcp_twsk_unique, 1710 .twsk_destructor= tcp_twsk_destructor, 1711 }; 1712 1713 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) 1714 { 1715 struct dst_entry *dst = skb_dst(skb); 1716 1717 if (dst) { 1718 dst_hold(dst); 1719 sk->sk_rx_dst = dst; 1720 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; 1721 } 1722 } 1723 EXPORT_SYMBOL(inet_sk_rx_dst_set); 1724 1725 const struct inet_connection_sock_af_ops ipv4_specific = { 1726 .queue_xmit = ip_queue_xmit, 1727 .send_check = tcp_v4_send_check, 1728 .rebuild_header = inet_sk_rebuild_header, 1729 .sk_rx_dst_set = inet_sk_rx_dst_set, 1730 .conn_request = tcp_v4_conn_request, 1731 .syn_recv_sock = tcp_v4_syn_recv_sock, 1732 .net_header_len = sizeof(struct iphdr), 1733 .setsockopt = ip_setsockopt, 1734 .getsockopt = ip_getsockopt, 1735 .addr2sockaddr = inet_csk_addr2sockaddr, 1736 .sockaddr_len = sizeof(struct sockaddr_in), 1737 .bind_conflict = inet_csk_bind_conflict, 1738 #ifdef CONFIG_COMPAT 1739 .compat_setsockopt = compat_ip_setsockopt, 1740 .compat_getsockopt = compat_ip_getsockopt, 1741 #endif 1742 .mtu_reduced = tcp_v4_mtu_reduced, 1743 }; 1744 EXPORT_SYMBOL(ipv4_specific); 1745 1746 #ifdef CONFIG_TCP_MD5SIG 1747 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { 1748 .md5_lookup = tcp_v4_md5_lookup, 1749 .calc_md5_hash = tcp_v4_md5_hash_skb, 1750 .md5_parse = tcp_v4_parse_md5_keys, 1751 }; 1752 #endif 1753 1754 /* NOTE: A lot of things set to zero explicitly by call to 1755 * sk_alloc() so need not be done here. 1756 */ 1757 static int tcp_v4_init_sock(struct sock *sk) 1758 { 1759 struct inet_connection_sock *icsk = inet_csk(sk); 1760 1761 tcp_init_sock(sk); 1762 1763 icsk->icsk_af_ops = &ipv4_specific; 1764 1765 #ifdef CONFIG_TCP_MD5SIG 1766 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; 1767 #endif 1768 1769 return 0; 1770 } 1771 1772 void tcp_v4_destroy_sock(struct sock *sk) 1773 { 1774 struct tcp_sock *tp = tcp_sk(sk); 1775 1776 tcp_clear_xmit_timers(sk); 1777 1778 tcp_cleanup_congestion_control(sk); 1779 1780 /* Cleanup up the write buffer. */ 1781 tcp_write_queue_purge(sk); 1782 1783 /* Cleans up our, hopefully empty, out_of_order_queue. */ 1784 __skb_queue_purge(&tp->out_of_order_queue); 1785 1786 #ifdef CONFIG_TCP_MD5SIG 1787 /* Clean up the MD5 key list, if any */ 1788 if (tp->md5sig_info) { 1789 tcp_clear_md5_list(sk); 1790 kfree_rcu(tp->md5sig_info, rcu); 1791 tp->md5sig_info = NULL; 1792 } 1793 #endif 1794 1795 /* Clean prequeue, it must be empty really */ 1796 __skb_queue_purge(&tp->ucopy.prequeue); 1797 1798 /* Clean up a referenced TCP bind bucket. */ 1799 if (inet_csk(sk)->icsk_bind_hash) 1800 inet_put_port(sk); 1801 1802 BUG_ON(tp->fastopen_rsk); 1803 1804 /* If socket is aborted during connect operation */ 1805 tcp_free_fastopen_req(tp); 1806 tcp_saved_syn_free(tp); 1807 1808 sk_sockets_allocated_dec(sk); 1809 sock_release_memcg(sk); 1810 } 1811 EXPORT_SYMBOL(tcp_v4_destroy_sock); 1812 1813 #ifdef CONFIG_PROC_FS 1814 /* Proc filesystem TCP sock list dumping. */ 1815 1816 /* 1817 * Get next listener socket follow cur. If cur is NULL, get first socket 1818 * starting from bucket given in st->bucket; when st->bucket is zero the 1819 * very first socket in the hash table is returned. 1820 */ 1821 static void *listening_get_next(struct seq_file *seq, void *cur) 1822 { 1823 struct inet_connection_sock *icsk; 1824 struct hlist_nulls_node *node; 1825 struct sock *sk = cur; 1826 struct inet_listen_hashbucket *ilb; 1827 struct tcp_iter_state *st = seq->private; 1828 struct net *net = seq_file_net(seq); 1829 1830 if (!sk) { 1831 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1832 spin_lock_bh(&ilb->lock); 1833 sk = sk_nulls_head(&ilb->head); 1834 st->offset = 0; 1835 goto get_sk; 1836 } 1837 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1838 ++st->num; 1839 ++st->offset; 1840 1841 if (st->state == TCP_SEQ_STATE_OPENREQ) { 1842 struct request_sock *req = cur; 1843 1844 icsk = inet_csk(st->syn_wait_sk); 1845 req = req->dl_next; 1846 while (1) { 1847 while (req) { 1848 if (req->rsk_ops->family == st->family) { 1849 cur = req; 1850 goto out; 1851 } 1852 req = req->dl_next; 1853 } 1854 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) 1855 break; 1856 get_req: 1857 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; 1858 } 1859 sk = sk_nulls_next(st->syn_wait_sk); 1860 st->state = TCP_SEQ_STATE_LISTENING; 1861 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1862 } else { 1863 icsk = inet_csk(sk); 1864 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1865 if (reqsk_queue_len(&icsk->icsk_accept_queue)) 1866 goto start_req; 1867 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1868 sk = sk_nulls_next(sk); 1869 } 1870 get_sk: 1871 sk_nulls_for_each_from(sk, node) { 1872 if (!net_eq(sock_net(sk), net)) 1873 continue; 1874 if (sk->sk_family == st->family) { 1875 cur = sk; 1876 goto out; 1877 } 1878 icsk = inet_csk(sk); 1879 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1880 if (reqsk_queue_len(&icsk->icsk_accept_queue)) { 1881 start_req: 1882 st->uid = sock_i_uid(sk); 1883 st->syn_wait_sk = sk; 1884 st->state = TCP_SEQ_STATE_OPENREQ; 1885 st->sbucket = 0; 1886 goto get_req; 1887 } 1888 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 1889 } 1890 spin_unlock_bh(&ilb->lock); 1891 st->offset = 0; 1892 if (++st->bucket < INET_LHTABLE_SIZE) { 1893 ilb = &tcp_hashinfo.listening_hash[st->bucket]; 1894 spin_lock_bh(&ilb->lock); 1895 sk = sk_nulls_head(&ilb->head); 1896 goto get_sk; 1897 } 1898 cur = NULL; 1899 out: 1900 return cur; 1901 } 1902 1903 static void *listening_get_idx(struct seq_file *seq, loff_t *pos) 1904 { 1905 struct tcp_iter_state *st = seq->private; 1906 void *rc; 1907 1908 st->bucket = 0; 1909 st->offset = 0; 1910 rc = listening_get_next(seq, NULL); 1911 1912 while (rc && *pos) { 1913 rc = listening_get_next(seq, rc); 1914 --*pos; 1915 } 1916 return rc; 1917 } 1918 1919 static inline bool empty_bucket(const struct tcp_iter_state *st) 1920 { 1921 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); 1922 } 1923 1924 /* 1925 * Get first established socket starting from bucket given in st->bucket. 1926 * If st->bucket is zero, the very first socket in the hash is returned. 1927 */ 1928 static void *established_get_first(struct seq_file *seq) 1929 { 1930 struct tcp_iter_state *st = seq->private; 1931 struct net *net = seq_file_net(seq); 1932 void *rc = NULL; 1933 1934 st->offset = 0; 1935 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { 1936 struct sock *sk; 1937 struct hlist_nulls_node *node; 1938 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); 1939 1940 /* Lockless fast path for the common case of empty buckets */ 1941 if (empty_bucket(st)) 1942 continue; 1943 1944 spin_lock_bh(lock); 1945 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { 1946 if (sk->sk_family != st->family || 1947 !net_eq(sock_net(sk), net)) { 1948 continue; 1949 } 1950 rc = sk; 1951 goto out; 1952 } 1953 spin_unlock_bh(lock); 1954 } 1955 out: 1956 return rc; 1957 } 1958 1959 static void *established_get_next(struct seq_file *seq, void *cur) 1960 { 1961 struct sock *sk = cur; 1962 struct hlist_nulls_node *node; 1963 struct tcp_iter_state *st = seq->private; 1964 struct net *net = seq_file_net(seq); 1965 1966 ++st->num; 1967 ++st->offset; 1968 1969 sk = sk_nulls_next(sk); 1970 1971 sk_nulls_for_each_from(sk, node) { 1972 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) 1973 return sk; 1974 } 1975 1976 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); 1977 ++st->bucket; 1978 return established_get_first(seq); 1979 } 1980 1981 static void *established_get_idx(struct seq_file *seq, loff_t pos) 1982 { 1983 struct tcp_iter_state *st = seq->private; 1984 void *rc; 1985 1986 st->bucket = 0; 1987 rc = established_get_first(seq); 1988 1989 while (rc && pos) { 1990 rc = established_get_next(seq, rc); 1991 --pos; 1992 } 1993 return rc; 1994 } 1995 1996 static void *tcp_get_idx(struct seq_file *seq, loff_t pos) 1997 { 1998 void *rc; 1999 struct tcp_iter_state *st = seq->private; 2000 2001 st->state = TCP_SEQ_STATE_LISTENING; 2002 rc = listening_get_idx(seq, &pos); 2003 2004 if (!rc) { 2005 st->state = TCP_SEQ_STATE_ESTABLISHED; 2006 rc = established_get_idx(seq, pos); 2007 } 2008 2009 return rc; 2010 } 2011 2012 static void *tcp_seek_last_pos(struct seq_file *seq) 2013 { 2014 struct tcp_iter_state *st = seq->private; 2015 int offset = st->offset; 2016 int orig_num = st->num; 2017 void *rc = NULL; 2018 2019 switch (st->state) { 2020 case TCP_SEQ_STATE_OPENREQ: 2021 case TCP_SEQ_STATE_LISTENING: 2022 if (st->bucket >= INET_LHTABLE_SIZE) 2023 break; 2024 st->state = TCP_SEQ_STATE_LISTENING; 2025 rc = listening_get_next(seq, NULL); 2026 while (offset-- && rc) 2027 rc = listening_get_next(seq, rc); 2028 if (rc) 2029 break; 2030 st->bucket = 0; 2031 st->state = TCP_SEQ_STATE_ESTABLISHED; 2032 /* Fallthrough */ 2033 case TCP_SEQ_STATE_ESTABLISHED: 2034 if (st->bucket > tcp_hashinfo.ehash_mask) 2035 break; 2036 rc = established_get_first(seq); 2037 while (offset-- && rc) 2038 rc = established_get_next(seq, rc); 2039 } 2040 2041 st->num = orig_num; 2042 2043 return rc; 2044 } 2045 2046 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) 2047 { 2048 struct tcp_iter_state *st = seq->private; 2049 void *rc; 2050 2051 if (*pos && *pos == st->last_pos) { 2052 rc = tcp_seek_last_pos(seq); 2053 if (rc) 2054 goto out; 2055 } 2056 2057 st->state = TCP_SEQ_STATE_LISTENING; 2058 st->num = 0; 2059 st->bucket = 0; 2060 st->offset = 0; 2061 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; 2062 2063 out: 2064 st->last_pos = *pos; 2065 return rc; 2066 } 2067 2068 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2069 { 2070 struct tcp_iter_state *st = seq->private; 2071 void *rc = NULL; 2072 2073 if (v == SEQ_START_TOKEN) { 2074 rc = tcp_get_idx(seq, 0); 2075 goto out; 2076 } 2077 2078 switch (st->state) { 2079 case TCP_SEQ_STATE_OPENREQ: 2080 case TCP_SEQ_STATE_LISTENING: 2081 rc = listening_get_next(seq, v); 2082 if (!rc) { 2083 st->state = TCP_SEQ_STATE_ESTABLISHED; 2084 st->bucket = 0; 2085 st->offset = 0; 2086 rc = established_get_first(seq); 2087 } 2088 break; 2089 case TCP_SEQ_STATE_ESTABLISHED: 2090 rc = established_get_next(seq, v); 2091 break; 2092 } 2093 out: 2094 ++*pos; 2095 st->last_pos = *pos; 2096 return rc; 2097 } 2098 2099 static void tcp_seq_stop(struct seq_file *seq, void *v) 2100 { 2101 struct tcp_iter_state *st = seq->private; 2102 2103 switch (st->state) { 2104 case TCP_SEQ_STATE_OPENREQ: 2105 if (v) { 2106 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); 2107 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); 2108 } 2109 case TCP_SEQ_STATE_LISTENING: 2110 if (v != SEQ_START_TOKEN) 2111 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); 2112 break; 2113 case TCP_SEQ_STATE_ESTABLISHED: 2114 if (v) 2115 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); 2116 break; 2117 } 2118 } 2119 2120 int tcp_seq_open(struct inode *inode, struct file *file) 2121 { 2122 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode); 2123 struct tcp_iter_state *s; 2124 int err; 2125 2126 err = seq_open_net(inode, file, &afinfo->seq_ops, 2127 sizeof(struct tcp_iter_state)); 2128 if (err < 0) 2129 return err; 2130 2131 s = ((struct seq_file *)file->private_data)->private; 2132 s->family = afinfo->family; 2133 s->last_pos = 0; 2134 return 0; 2135 } 2136 EXPORT_SYMBOL(tcp_seq_open); 2137 2138 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) 2139 { 2140 int rc = 0; 2141 struct proc_dir_entry *p; 2142 2143 afinfo->seq_ops.start = tcp_seq_start; 2144 afinfo->seq_ops.next = tcp_seq_next; 2145 afinfo->seq_ops.stop = tcp_seq_stop; 2146 2147 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2148 afinfo->seq_fops, afinfo); 2149 if (!p) 2150 rc = -ENOMEM; 2151 return rc; 2152 } 2153 EXPORT_SYMBOL(tcp_proc_register); 2154 2155 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) 2156 { 2157 remove_proc_entry(afinfo->name, net->proc_net); 2158 } 2159 EXPORT_SYMBOL(tcp_proc_unregister); 2160 2161 static void get_openreq4(const struct request_sock *req, 2162 struct seq_file *f, int i, kuid_t uid) 2163 { 2164 const struct inet_request_sock *ireq = inet_rsk(req); 2165 long delta = req->rsk_timer.expires - jiffies; 2166 2167 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 2168 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", 2169 i, 2170 ireq->ir_loc_addr, 2171 ireq->ir_num, 2172 ireq->ir_rmt_addr, 2173 ntohs(ireq->ir_rmt_port), 2174 TCP_SYN_RECV, 2175 0, 0, /* could print option size, but that is af dependent. */ 2176 1, /* timers active (only the expire timer) */ 2177 jiffies_delta_to_clock_t(delta), 2178 req->num_timeout, 2179 from_kuid_munged(seq_user_ns(f), uid), 2180 0, /* non standard timer */ 2181 0, /* open_requests have no inode */ 2182 0, 2183 req); 2184 } 2185 2186 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) 2187 { 2188 int timer_active; 2189 unsigned long timer_expires; 2190 const struct tcp_sock *tp = tcp_sk(sk); 2191 const struct inet_connection_sock *icsk = inet_csk(sk); 2192 const struct inet_sock *inet = inet_sk(sk); 2193 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq; 2194 __be32 dest = inet->inet_daddr; 2195 __be32 src = inet->inet_rcv_saddr; 2196 __u16 destp = ntohs(inet->inet_dport); 2197 __u16 srcp = ntohs(inet->inet_sport); 2198 int rx_queue; 2199 2200 if (icsk->icsk_pending == ICSK_TIME_RETRANS || 2201 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || 2202 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { 2203 timer_active = 1; 2204 timer_expires = icsk->icsk_timeout; 2205 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { 2206 timer_active = 4; 2207 timer_expires = icsk->icsk_timeout; 2208 } else if (timer_pending(&sk->sk_timer)) { 2209 timer_active = 2; 2210 timer_expires = sk->sk_timer.expires; 2211 } else { 2212 timer_active = 0; 2213 timer_expires = jiffies; 2214 } 2215 2216 if (sk->sk_state == TCP_LISTEN) 2217 rx_queue = sk->sk_ack_backlog; 2218 else 2219 /* 2220 * because we dont lock socket, we might find a transient negative value 2221 */ 2222 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); 2223 2224 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " 2225 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", 2226 i, src, srcp, dest, destp, sk->sk_state, 2227 tp->write_seq - tp->snd_una, 2228 rx_queue, 2229 timer_active, 2230 jiffies_delta_to_clock_t(timer_expires - jiffies), 2231 icsk->icsk_retransmits, 2232 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), 2233 icsk->icsk_probes_out, 2234 sock_i_ino(sk), 2235 atomic_read(&sk->sk_refcnt), sk, 2236 jiffies_to_clock_t(icsk->icsk_rto), 2237 jiffies_to_clock_t(icsk->icsk_ack.ato), 2238 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, 2239 tp->snd_cwnd, 2240 sk->sk_state == TCP_LISTEN ? 2241 (fastopenq ? fastopenq->max_qlen : 0) : 2242 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); 2243 } 2244 2245 static void get_timewait4_sock(const struct inet_timewait_sock *tw, 2246 struct seq_file *f, int i) 2247 { 2248 long delta = tw->tw_timer.expires - jiffies; 2249 __be32 dest, src; 2250 __u16 destp, srcp; 2251 2252 dest = tw->tw_daddr; 2253 src = tw->tw_rcv_saddr; 2254 destp = ntohs(tw->tw_dport); 2255 srcp = ntohs(tw->tw_sport); 2256 2257 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 2258 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", 2259 i, src, srcp, dest, destp, tw->tw_substate, 0, 0, 2260 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, 2261 atomic_read(&tw->tw_refcnt), tw); 2262 } 2263 2264 #define TMPSZ 150 2265 2266 static int tcp4_seq_show(struct seq_file *seq, void *v) 2267 { 2268 struct tcp_iter_state *st; 2269 struct sock *sk = v; 2270 2271 seq_setwidth(seq, TMPSZ - 1); 2272 if (v == SEQ_START_TOKEN) { 2273 seq_puts(seq, " sl local_address rem_address st tx_queue " 2274 "rx_queue tr tm->when retrnsmt uid timeout " 2275 "inode"); 2276 goto out; 2277 } 2278 st = seq->private; 2279 2280 switch (st->state) { 2281 case TCP_SEQ_STATE_LISTENING: 2282 case TCP_SEQ_STATE_ESTABLISHED: 2283 if (sk->sk_state == TCP_TIME_WAIT) 2284 get_timewait4_sock(v, seq, st->num); 2285 else 2286 get_tcp4_sock(v, seq, st->num); 2287 break; 2288 case TCP_SEQ_STATE_OPENREQ: 2289 get_openreq4(v, seq, st->num, st->uid); 2290 break; 2291 } 2292 out: 2293 seq_pad(seq, '\n'); 2294 return 0; 2295 } 2296 2297 static const struct file_operations tcp_afinfo_seq_fops = { 2298 .owner = THIS_MODULE, 2299 .open = tcp_seq_open, 2300 .read = seq_read, 2301 .llseek = seq_lseek, 2302 .release = seq_release_net 2303 }; 2304 2305 static struct tcp_seq_afinfo tcp4_seq_afinfo = { 2306 .name = "tcp", 2307 .family = AF_INET, 2308 .seq_fops = &tcp_afinfo_seq_fops, 2309 .seq_ops = { 2310 .show = tcp4_seq_show, 2311 }, 2312 }; 2313 2314 static int __net_init tcp4_proc_init_net(struct net *net) 2315 { 2316 return tcp_proc_register(net, &tcp4_seq_afinfo); 2317 } 2318 2319 static void __net_exit tcp4_proc_exit_net(struct net *net) 2320 { 2321 tcp_proc_unregister(net, &tcp4_seq_afinfo); 2322 } 2323 2324 static struct pernet_operations tcp4_net_ops = { 2325 .init = tcp4_proc_init_net, 2326 .exit = tcp4_proc_exit_net, 2327 }; 2328 2329 int __init tcp4_proc_init(void) 2330 { 2331 return register_pernet_subsys(&tcp4_net_ops); 2332 } 2333 2334 void tcp4_proc_exit(void) 2335 { 2336 unregister_pernet_subsys(&tcp4_net_ops); 2337 } 2338 #endif /* CONFIG_PROC_FS */ 2339 2340 struct proto tcp_prot = { 2341 .name = "TCP", 2342 .owner = THIS_MODULE, 2343 .close = tcp_close, 2344 .connect = tcp_v4_connect, 2345 .disconnect = tcp_disconnect, 2346 .accept = inet_csk_accept, 2347 .ioctl = tcp_ioctl, 2348 .init = tcp_v4_init_sock, 2349 .destroy = tcp_v4_destroy_sock, 2350 .shutdown = tcp_shutdown, 2351 .setsockopt = tcp_setsockopt, 2352 .getsockopt = tcp_getsockopt, 2353 .recvmsg = tcp_recvmsg, 2354 .sendmsg = tcp_sendmsg, 2355 .sendpage = tcp_sendpage, 2356 .backlog_rcv = tcp_v4_do_rcv, 2357 .release_cb = tcp_release_cb, 2358 .hash = inet_hash, 2359 .unhash = inet_unhash, 2360 .get_port = inet_csk_get_port, 2361 .enter_memory_pressure = tcp_enter_memory_pressure, 2362 .stream_memory_free = tcp_stream_memory_free, 2363 .sockets_allocated = &tcp_sockets_allocated, 2364 .orphan_count = &tcp_orphan_count, 2365 .memory_allocated = &tcp_memory_allocated, 2366 .memory_pressure = &tcp_memory_pressure, 2367 .sysctl_mem = sysctl_tcp_mem, 2368 .sysctl_wmem = sysctl_tcp_wmem, 2369 .sysctl_rmem = sysctl_tcp_rmem, 2370 .max_header = MAX_TCP_HEADER, 2371 .obj_size = sizeof(struct tcp_sock), 2372 .slab_flags = SLAB_DESTROY_BY_RCU, 2373 .twsk_prot = &tcp_timewait_sock_ops, 2374 .rsk_prot = &tcp_request_sock_ops, 2375 .h.hashinfo = &tcp_hashinfo, 2376 .no_autobind = true, 2377 #ifdef CONFIG_COMPAT 2378 .compat_setsockopt = compat_tcp_setsockopt, 2379 .compat_getsockopt = compat_tcp_getsockopt, 2380 #endif 2381 #ifdef CONFIG_MEMCG_KMEM 2382 .init_cgroup = tcp_init_cgroup, 2383 .destroy_cgroup = tcp_destroy_cgroup, 2384 .proto_cgroup = tcp_proto_cgroup, 2385 #endif 2386 }; 2387 EXPORT_SYMBOL(tcp_prot); 2388 2389 static void __net_exit tcp_sk_exit(struct net *net) 2390 { 2391 int cpu; 2392 2393 for_each_possible_cpu(cpu) 2394 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu)); 2395 free_percpu(net->ipv4.tcp_sk); 2396 } 2397 2398 static int __net_init tcp_sk_init(struct net *net) 2399 { 2400 int res, cpu; 2401 2402 net->ipv4.tcp_sk = alloc_percpu(struct sock *); 2403 if (!net->ipv4.tcp_sk) 2404 return -ENOMEM; 2405 2406 for_each_possible_cpu(cpu) { 2407 struct sock *sk; 2408 2409 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW, 2410 IPPROTO_TCP, net); 2411 if (res) 2412 goto fail; 2413 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk; 2414 } 2415 2416 net->ipv4.sysctl_tcp_ecn = 2; 2417 net->ipv4.sysctl_tcp_ecn_fallback = 1; 2418 2419 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; 2420 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD; 2421 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL; 2422 2423 return 0; 2424 fail: 2425 tcp_sk_exit(net); 2426 2427 return res; 2428 } 2429 2430 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) 2431 { 2432 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); 2433 } 2434 2435 static struct pernet_operations __net_initdata tcp_sk_ops = { 2436 .init = tcp_sk_init, 2437 .exit = tcp_sk_exit, 2438 .exit_batch = tcp_sk_exit_batch, 2439 }; 2440 2441 void __init tcp_v4_init(void) 2442 { 2443 inet_hashinfo_init(&tcp_hashinfo); 2444 if (register_pernet_subsys(&tcp_sk_ops)) 2445 panic("Failed to create the TCP control socket.\n"); 2446 } 2447