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 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Mark Evans, <evansmp@uhura.aston.ac.uk> 11 * Corey Minyard <wf-rch!minyard@relay.EU.net> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> 14 * Linus Torvalds, <torvalds@cs.helsinki.fi> 15 * Alan Cox, <gw4pts@gw4pts.ampr.org> 16 * Matthew Dillon, <dillon@apollo.west.oic.com> 17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 18 * Jorge Cwik, <jorge@laser.satlink.net> 19 */ 20 21 #include <linux/mm.h> 22 #include <linux/module.h> 23 #include <linux/slab.h> 24 #include <linux/sysctl.h> 25 #include <linux/workqueue.h> 26 #include <net/tcp.h> 27 #include <net/inet_common.h> 28 #include <net/xfrm.h> 29 30 int sysctl_tcp_syncookies __read_mostly = 1; 31 EXPORT_SYMBOL(sysctl_tcp_syncookies); 32 33 int sysctl_tcp_abort_on_overflow __read_mostly; 34 35 struct inet_timewait_death_row tcp_death_row = { 36 .sysctl_max_tw_buckets = NR_FILE * 2, 37 .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS, 38 .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock), 39 .hashinfo = &tcp_hashinfo, 40 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0, 41 (unsigned long)&tcp_death_row), 42 .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work, 43 inet_twdr_twkill_work), 44 /* Short-time timewait calendar */ 45 46 .twcal_hand = -1, 47 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0, 48 (unsigned long)&tcp_death_row), 49 }; 50 EXPORT_SYMBOL_GPL(tcp_death_row); 51 52 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) 53 { 54 if (seq == s_win) 55 return true; 56 if (after(end_seq, s_win) && before(seq, e_win)) 57 return true; 58 return seq == e_win && seq == end_seq; 59 } 60 61 /* 62 * * Main purpose of TIME-WAIT state is to close connection gracefully, 63 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN 64 * (and, probably, tail of data) and one or more our ACKs are lost. 65 * * What is TIME-WAIT timeout? It is associated with maximal packet 66 * lifetime in the internet, which results in wrong conclusion, that 67 * it is set to catch "old duplicate segments" wandering out of their path. 68 * It is not quite correct. This timeout is calculated so that it exceeds 69 * maximal retransmission timeout enough to allow to lose one (or more) 70 * segments sent by peer and our ACKs. This time may be calculated from RTO. 71 * * When TIME-WAIT socket receives RST, it means that another end 72 * finally closed and we are allowed to kill TIME-WAIT too. 73 * * Second purpose of TIME-WAIT is catching old duplicate segments. 74 * Well, certainly it is pure paranoia, but if we load TIME-WAIT 75 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. 76 * * If we invented some more clever way to catch duplicates 77 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. 78 * 79 * The algorithm below is based on FORMAL INTERPRETATION of RFCs. 80 * When you compare it to RFCs, please, read section SEGMENT ARRIVES 81 * from the very beginning. 82 * 83 * NOTE. With recycling (and later with fin-wait-2) TW bucket 84 * is _not_ stateless. It means, that strictly speaking we must 85 * spinlock it. I do not want! Well, probability of misbehaviour 86 * is ridiculously low and, seems, we could use some mb() tricks 87 * to avoid misread sequence numbers, states etc. --ANK 88 * 89 * We don't need to initialize tmp_out.sack_ok as we don't use the results 90 */ 91 enum tcp_tw_status 92 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, 93 const struct tcphdr *th) 94 { 95 struct tcp_options_received tmp_opt; 96 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); 97 bool paws_reject = false; 98 99 tmp_opt.saw_tstamp = 0; 100 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { 101 tcp_parse_options(skb, &tmp_opt, 0, NULL); 102 103 if (tmp_opt.saw_tstamp) { 104 tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset; 105 tmp_opt.ts_recent = tcptw->tw_ts_recent; 106 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; 107 paws_reject = tcp_paws_reject(&tmp_opt, th->rst); 108 } 109 } 110 111 if (tw->tw_substate == TCP_FIN_WAIT2) { 112 /* Just repeat all the checks of tcp_rcv_state_process() */ 113 114 /* Out of window, send ACK */ 115 if (paws_reject || 116 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, 117 tcptw->tw_rcv_nxt, 118 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) 119 return TCP_TW_ACK; 120 121 if (th->rst) 122 goto kill; 123 124 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) 125 goto kill_with_rst; 126 127 /* Dup ACK? */ 128 if (!th->ack || 129 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || 130 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { 131 inet_twsk_put(tw); 132 return TCP_TW_SUCCESS; 133 } 134 135 /* New data or FIN. If new data arrive after half-duplex close, 136 * reset. 137 */ 138 if (!th->fin || 139 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) { 140 kill_with_rst: 141 inet_twsk_deschedule(tw, &tcp_death_row); 142 inet_twsk_put(tw); 143 return TCP_TW_RST; 144 } 145 146 /* FIN arrived, enter true time-wait state. */ 147 tw->tw_substate = TCP_TIME_WAIT; 148 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; 149 if (tmp_opt.saw_tstamp) { 150 tcptw->tw_ts_recent_stamp = get_seconds(); 151 tcptw->tw_ts_recent = tmp_opt.rcv_tsval; 152 } 153 154 if (tcp_death_row.sysctl_tw_recycle && 155 tcptw->tw_ts_recent_stamp && 156 tcp_tw_remember_stamp(tw)) 157 inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout, 158 TCP_TIMEWAIT_LEN); 159 else 160 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 161 TCP_TIMEWAIT_LEN); 162 return TCP_TW_ACK; 163 } 164 165 /* 166 * Now real TIME-WAIT state. 167 * 168 * RFC 1122: 169 * "When a connection is [...] on TIME-WAIT state [...] 170 * [a TCP] MAY accept a new SYN from the remote TCP to 171 * reopen the connection directly, if it: 172 * 173 * (1) assigns its initial sequence number for the new 174 * connection to be larger than the largest sequence 175 * number it used on the previous connection incarnation, 176 * and 177 * 178 * (2) returns to TIME-WAIT state if the SYN turns out 179 * to be an old duplicate". 180 */ 181 182 if (!paws_reject && 183 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && 184 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { 185 /* In window segment, it may be only reset or bare ack. */ 186 187 if (th->rst) { 188 /* This is TIME_WAIT assassination, in two flavors. 189 * Oh well... nobody has a sufficient solution to this 190 * protocol bug yet. 191 */ 192 if (sysctl_tcp_rfc1337 == 0) { 193 kill: 194 inet_twsk_deschedule(tw, &tcp_death_row); 195 inet_twsk_put(tw); 196 return TCP_TW_SUCCESS; 197 } 198 } 199 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 200 TCP_TIMEWAIT_LEN); 201 202 if (tmp_opt.saw_tstamp) { 203 tcptw->tw_ts_recent = tmp_opt.rcv_tsval; 204 tcptw->tw_ts_recent_stamp = get_seconds(); 205 } 206 207 inet_twsk_put(tw); 208 return TCP_TW_SUCCESS; 209 } 210 211 /* Out of window segment. 212 213 All the segments are ACKed immediately. 214 215 The only exception is new SYN. We accept it, if it is 216 not old duplicate and we are not in danger to be killed 217 by delayed old duplicates. RFC check is that it has 218 newer sequence number works at rates <40Mbit/sec. 219 However, if paws works, it is reliable AND even more, 220 we even may relax silly seq space cutoff. 221 222 RED-PEN: we violate main RFC requirement, if this SYN will appear 223 old duplicate (i.e. we receive RST in reply to SYN-ACK), 224 we must return socket to time-wait state. It is not good, 225 but not fatal yet. 226 */ 227 228 if (th->syn && !th->rst && !th->ack && !paws_reject && 229 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || 230 (tmp_opt.saw_tstamp && 231 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { 232 u32 isn = tcptw->tw_snd_nxt + 65535 + 2; 233 if (isn == 0) 234 isn++; 235 TCP_SKB_CB(skb)->when = isn; 236 return TCP_TW_SYN; 237 } 238 239 if (paws_reject) 240 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED); 241 242 if (!th->rst) { 243 /* In this case we must reset the TIMEWAIT timer. 244 * 245 * If it is ACKless SYN it may be both old duplicate 246 * and new good SYN with random sequence number <rcv_nxt. 247 * Do not reschedule in the last case. 248 */ 249 if (paws_reject || th->ack) 250 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 251 TCP_TIMEWAIT_LEN); 252 253 /* Send ACK. Note, we do not put the bucket, 254 * it will be released by caller. 255 */ 256 return TCP_TW_ACK; 257 } 258 inet_twsk_put(tw); 259 return TCP_TW_SUCCESS; 260 } 261 EXPORT_SYMBOL(tcp_timewait_state_process); 262 263 /* 264 * Move a socket to time-wait or dead fin-wait-2 state. 265 */ 266 void tcp_time_wait(struct sock *sk, int state, int timeo) 267 { 268 struct inet_timewait_sock *tw = NULL; 269 const struct inet_connection_sock *icsk = inet_csk(sk); 270 const struct tcp_sock *tp = tcp_sk(sk); 271 bool recycle_ok = false; 272 273 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp) 274 recycle_ok = tcp_remember_stamp(sk); 275 276 if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets) 277 tw = inet_twsk_alloc(sk, state); 278 279 if (tw != NULL) { 280 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); 281 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); 282 struct inet_sock *inet = inet_sk(sk); 283 284 tw->tw_transparent = inet->transparent; 285 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; 286 tcptw->tw_rcv_nxt = tp->rcv_nxt; 287 tcptw->tw_snd_nxt = tp->snd_nxt; 288 tcptw->tw_rcv_wnd = tcp_receive_window(tp); 289 tcptw->tw_ts_recent = tp->rx_opt.ts_recent; 290 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; 291 tcptw->tw_ts_offset = tp->tsoffset; 292 293 #if IS_ENABLED(CONFIG_IPV6) 294 if (tw->tw_family == PF_INET6) { 295 struct ipv6_pinfo *np = inet6_sk(sk); 296 297 tw->tw_v6_daddr = sk->sk_v6_daddr; 298 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 299 tw->tw_tclass = np->tclass; 300 tw->tw_flowlabel = np->flow_label >> 12; 301 tw->tw_ipv6only = sk->sk_ipv6only; 302 } 303 #endif 304 305 #ifdef CONFIG_TCP_MD5SIG 306 /* 307 * The timewait bucket does not have the key DB from the 308 * sock structure. We just make a quick copy of the 309 * md5 key being used (if indeed we are using one) 310 * so the timewait ack generating code has the key. 311 */ 312 do { 313 struct tcp_md5sig_key *key; 314 tcptw->tw_md5_key = NULL; 315 key = tp->af_specific->md5_lookup(sk, sk); 316 if (key != NULL) { 317 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC); 318 if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool()) 319 BUG(); 320 } 321 } while (0); 322 #endif 323 324 /* Linkage updates. */ 325 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo); 326 327 /* Get the TIME_WAIT timeout firing. */ 328 if (timeo < rto) 329 timeo = rto; 330 331 if (recycle_ok) { 332 tw->tw_timeout = rto; 333 } else { 334 tw->tw_timeout = TCP_TIMEWAIT_LEN; 335 if (state == TCP_TIME_WAIT) 336 timeo = TCP_TIMEWAIT_LEN; 337 } 338 339 inet_twsk_schedule(tw, &tcp_death_row, timeo, 340 TCP_TIMEWAIT_LEN); 341 inet_twsk_put(tw); 342 } else { 343 /* Sorry, if we're out of memory, just CLOSE this 344 * socket up. We've got bigger problems than 345 * non-graceful socket closings. 346 */ 347 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW); 348 } 349 350 tcp_update_metrics(sk); 351 tcp_done(sk); 352 } 353 354 void tcp_twsk_destructor(struct sock *sk) 355 { 356 #ifdef CONFIG_TCP_MD5SIG 357 struct tcp_timewait_sock *twsk = tcp_twsk(sk); 358 359 if (twsk->tw_md5_key) 360 kfree_rcu(twsk->tw_md5_key, rcu); 361 #endif 362 } 363 EXPORT_SYMBOL_GPL(tcp_twsk_destructor); 364 365 void tcp_openreq_init_rwin(struct request_sock *req, 366 struct sock *sk, struct dst_entry *dst) 367 { 368 struct inet_request_sock *ireq = inet_rsk(req); 369 struct tcp_sock *tp = tcp_sk(sk); 370 __u8 rcv_wscale; 371 int mss = dst_metric_advmss(dst); 372 373 if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss) 374 mss = tp->rx_opt.user_mss; 375 376 /* Set this up on the first call only */ 377 req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW); 378 379 /* limit the window selection if the user enforce a smaller rx buffer */ 380 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && 381 (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0)) 382 req->window_clamp = tcp_full_space(sk); 383 384 /* tcp_full_space because it is guaranteed to be the first packet */ 385 tcp_select_initial_window(tcp_full_space(sk), 386 mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0), 387 &req->rcv_wnd, 388 &req->window_clamp, 389 ireq->wscale_ok, 390 &rcv_wscale, 391 dst_metric(dst, RTAX_INITRWND)); 392 ireq->rcv_wscale = rcv_wscale; 393 } 394 EXPORT_SYMBOL(tcp_openreq_init_rwin); 395 396 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp, 397 struct request_sock *req) 398 { 399 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0; 400 } 401 402 /* This is not only more efficient than what we used to do, it eliminates 403 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM 404 * 405 * Actually, we could lots of memory writes here. tp of listening 406 * socket contains all necessary default parameters. 407 */ 408 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb) 409 { 410 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC); 411 412 if (newsk != NULL) { 413 const struct inet_request_sock *ireq = inet_rsk(req); 414 struct tcp_request_sock *treq = tcp_rsk(req); 415 struct inet_connection_sock *newicsk = inet_csk(newsk); 416 struct tcp_sock *newtp = tcp_sk(newsk); 417 418 /* Now setup tcp_sock */ 419 newtp->pred_flags = 0; 420 421 newtp->rcv_wup = newtp->copied_seq = 422 newtp->rcv_nxt = treq->rcv_isn + 1; 423 424 newtp->snd_sml = newtp->snd_una = 425 newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1; 426 427 tcp_prequeue_init(newtp); 428 INIT_LIST_HEAD(&newtp->tsq_node); 429 430 tcp_init_wl(newtp, treq->rcv_isn); 431 432 newtp->srtt_us = 0; 433 newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); 434 newicsk->icsk_rto = TCP_TIMEOUT_INIT; 435 436 newtp->packets_out = 0; 437 newtp->retrans_out = 0; 438 newtp->sacked_out = 0; 439 newtp->fackets_out = 0; 440 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH; 441 tcp_enable_early_retrans(newtp); 442 newtp->tlp_high_seq = 0; 443 newtp->lsndtime = treq->snt_synack; 444 newtp->total_retrans = req->num_retrans; 445 446 /* So many TCP implementations out there (incorrectly) count the 447 * initial SYN frame in their delayed-ACK and congestion control 448 * algorithms that we must have the following bandaid to talk 449 * efficiently to them. -DaveM 450 */ 451 newtp->snd_cwnd = TCP_INIT_CWND; 452 newtp->snd_cwnd_cnt = 0; 453 454 if (newicsk->icsk_ca_ops != &tcp_init_congestion_ops && 455 !try_module_get(newicsk->icsk_ca_ops->owner)) 456 newicsk->icsk_ca_ops = &tcp_init_congestion_ops; 457 458 tcp_set_ca_state(newsk, TCP_CA_Open); 459 tcp_init_xmit_timers(newsk); 460 __skb_queue_head_init(&newtp->out_of_order_queue); 461 newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1; 462 463 newtp->rx_opt.saw_tstamp = 0; 464 465 newtp->rx_opt.dsack = 0; 466 newtp->rx_opt.num_sacks = 0; 467 468 newtp->urg_data = 0; 469 470 if (sock_flag(newsk, SOCK_KEEPOPEN)) 471 inet_csk_reset_keepalive_timer(newsk, 472 keepalive_time_when(newtp)); 473 474 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; 475 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) { 476 if (sysctl_tcp_fack) 477 tcp_enable_fack(newtp); 478 } 479 newtp->window_clamp = req->window_clamp; 480 newtp->rcv_ssthresh = req->rcv_wnd; 481 newtp->rcv_wnd = req->rcv_wnd; 482 newtp->rx_opt.wscale_ok = ireq->wscale_ok; 483 if (newtp->rx_opt.wscale_ok) { 484 newtp->rx_opt.snd_wscale = ireq->snd_wscale; 485 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; 486 } else { 487 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; 488 newtp->window_clamp = min(newtp->window_clamp, 65535U); 489 } 490 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) << 491 newtp->rx_opt.snd_wscale); 492 newtp->max_window = newtp->snd_wnd; 493 494 if (newtp->rx_opt.tstamp_ok) { 495 newtp->rx_opt.ts_recent = req->ts_recent; 496 newtp->rx_opt.ts_recent_stamp = get_seconds(); 497 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; 498 } else { 499 newtp->rx_opt.ts_recent_stamp = 0; 500 newtp->tcp_header_len = sizeof(struct tcphdr); 501 } 502 newtp->tsoffset = 0; 503 #ifdef CONFIG_TCP_MD5SIG 504 newtp->md5sig_info = NULL; /*XXX*/ 505 if (newtp->af_specific->md5_lookup(sk, newsk)) 506 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; 507 #endif 508 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len) 509 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; 510 newtp->rx_opt.mss_clamp = req->mss; 511 TCP_ECN_openreq_child(newtp, req); 512 newtp->fastopen_rsk = NULL; 513 newtp->syn_data_acked = 0; 514 515 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS); 516 } 517 return newsk; 518 } 519 EXPORT_SYMBOL(tcp_create_openreq_child); 520 521 /* 522 * Process an incoming packet for SYN_RECV sockets represented as a 523 * request_sock. Normally sk is the listener socket but for TFO it 524 * points to the child socket. 525 * 526 * XXX (TFO) - The current impl contains a special check for ack 527 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better? 528 * 529 * We don't need to initialize tmp_opt.sack_ok as we don't use the results 530 */ 531 532 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 533 struct request_sock *req, 534 struct request_sock **prev, 535 bool fastopen) 536 { 537 struct tcp_options_received tmp_opt; 538 struct sock *child; 539 const struct tcphdr *th = tcp_hdr(skb); 540 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); 541 bool paws_reject = false; 542 543 BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN)); 544 545 tmp_opt.saw_tstamp = 0; 546 if (th->doff > (sizeof(struct tcphdr)>>2)) { 547 tcp_parse_options(skb, &tmp_opt, 0, NULL); 548 549 if (tmp_opt.saw_tstamp) { 550 tmp_opt.ts_recent = req->ts_recent; 551 /* We do not store true stamp, but it is not required, 552 * it can be estimated (approximately) 553 * from another data. 554 */ 555 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout); 556 paws_reject = tcp_paws_reject(&tmp_opt, th->rst); 557 } 558 } 559 560 /* Check for pure retransmitted SYN. */ 561 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && 562 flg == TCP_FLAG_SYN && 563 !paws_reject) { 564 /* 565 * RFC793 draws (Incorrectly! It was fixed in RFC1122) 566 * this case on figure 6 and figure 8, but formal 567 * protocol description says NOTHING. 568 * To be more exact, it says that we should send ACK, 569 * because this segment (at least, if it has no data) 570 * is out of window. 571 * 572 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT 573 * describe SYN-RECV state. All the description 574 * is wrong, we cannot believe to it and should 575 * rely only on common sense and implementation 576 * experience. 577 * 578 * Enforce "SYN-ACK" according to figure 8, figure 6 579 * of RFC793, fixed by RFC1122. 580 * 581 * Note that even if there is new data in the SYN packet 582 * they will be thrown away too. 583 * 584 * Reset timer after retransmitting SYNACK, similar to 585 * the idea of fast retransmit in recovery. 586 */ 587 if (!inet_rtx_syn_ack(sk, req)) 588 req->expires = min(TCP_TIMEOUT_INIT << req->num_timeout, 589 TCP_RTO_MAX) + jiffies; 590 return NULL; 591 } 592 593 /* Further reproduces section "SEGMENT ARRIVES" 594 for state SYN-RECEIVED of RFC793. 595 It is broken, however, it does not work only 596 when SYNs are crossed. 597 598 You would think that SYN crossing is impossible here, since 599 we should have a SYN_SENT socket (from connect()) on our end, 600 but this is not true if the crossed SYNs were sent to both 601 ends by a malicious third party. We must defend against this, 602 and to do that we first verify the ACK (as per RFC793, page 603 36) and reset if it is invalid. Is this a true full defense? 604 To convince ourselves, let us consider a way in which the ACK 605 test can still pass in this 'malicious crossed SYNs' case. 606 Malicious sender sends identical SYNs (and thus identical sequence 607 numbers) to both A and B: 608 609 A: gets SYN, seq=7 610 B: gets SYN, seq=7 611 612 By our good fortune, both A and B select the same initial 613 send sequence number of seven :-) 614 615 A: sends SYN|ACK, seq=7, ack_seq=8 616 B: sends SYN|ACK, seq=7, ack_seq=8 617 618 So we are now A eating this SYN|ACK, ACK test passes. So 619 does sequence test, SYN is truncated, and thus we consider 620 it a bare ACK. 621 622 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this 623 bare ACK. Otherwise, we create an established connection. Both 624 ends (listening sockets) accept the new incoming connection and try 625 to talk to each other. 8-) 626 627 Note: This case is both harmless, and rare. Possibility is about the 628 same as us discovering intelligent life on another plant tomorrow. 629 630 But generally, we should (RFC lies!) to accept ACK 631 from SYNACK both here and in tcp_rcv_state_process(). 632 tcp_rcv_state_process() does not, hence, we do not too. 633 634 Note that the case is absolutely generic: 635 we cannot optimize anything here without 636 violating protocol. All the checks must be made 637 before attempt to create socket. 638 */ 639 640 /* RFC793 page 36: "If the connection is in any non-synchronized state ... 641 * and the incoming segment acknowledges something not yet 642 * sent (the segment carries an unacceptable ACK) ... 643 * a reset is sent." 644 * 645 * Invalid ACK: reset will be sent by listening socket. 646 * Note that the ACK validity check for a Fast Open socket is done 647 * elsewhere and is checked directly against the child socket rather 648 * than req because user data may have been sent out. 649 */ 650 if ((flg & TCP_FLAG_ACK) && !fastopen && 651 (TCP_SKB_CB(skb)->ack_seq != 652 tcp_rsk(req)->snt_isn + 1)) 653 return sk; 654 655 /* Also, it would be not so bad idea to check rcv_tsecr, which 656 * is essentially ACK extension and too early or too late values 657 * should cause reset in unsynchronized states. 658 */ 659 660 /* RFC793: "first check sequence number". */ 661 662 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, 663 tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) { 664 /* Out of window: send ACK and drop. */ 665 if (!(flg & TCP_FLAG_RST)) 666 req->rsk_ops->send_ack(sk, skb, req); 667 if (paws_reject) 668 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); 669 return NULL; 670 } 671 672 /* In sequence, PAWS is OK. */ 673 674 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt)) 675 req->ts_recent = tmp_opt.rcv_tsval; 676 677 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { 678 /* Truncate SYN, it is out of window starting 679 at tcp_rsk(req)->rcv_isn + 1. */ 680 flg &= ~TCP_FLAG_SYN; 681 } 682 683 /* RFC793: "second check the RST bit" and 684 * "fourth, check the SYN bit" 685 */ 686 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { 687 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); 688 goto embryonic_reset; 689 } 690 691 /* ACK sequence verified above, just make sure ACK is 692 * set. If ACK not set, just silently drop the packet. 693 * 694 * XXX (TFO) - if we ever allow "data after SYN", the 695 * following check needs to be removed. 696 */ 697 if (!(flg & TCP_FLAG_ACK)) 698 return NULL; 699 700 /* For Fast Open no more processing is needed (sk is the 701 * child socket). 702 */ 703 if (fastopen) 704 return sk; 705 706 /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ 707 if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && 708 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { 709 inet_rsk(req)->acked = 1; 710 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP); 711 return NULL; 712 } 713 714 /* OK, ACK is valid, create big socket and 715 * feed this segment to it. It will repeat all 716 * the tests. THIS SEGMENT MUST MOVE SOCKET TO 717 * ESTABLISHED STATE. If it will be dropped after 718 * socket is created, wait for troubles. 719 */ 720 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL); 721 if (child == NULL) 722 goto listen_overflow; 723 724 inet_csk_reqsk_queue_unlink(sk, req, prev); 725 inet_csk_reqsk_queue_removed(sk, req); 726 727 inet_csk_reqsk_queue_add(sk, req, child); 728 return child; 729 730 listen_overflow: 731 if (!sysctl_tcp_abort_on_overflow) { 732 inet_rsk(req)->acked = 1; 733 return NULL; 734 } 735 736 embryonic_reset: 737 if (!(flg & TCP_FLAG_RST)) { 738 /* Received a bad SYN pkt - for TFO We try not to reset 739 * the local connection unless it's really necessary to 740 * avoid becoming vulnerable to outside attack aiming at 741 * resetting legit local connections. 742 */ 743 req->rsk_ops->send_reset(sk, skb); 744 } else if (fastopen) { /* received a valid RST pkt */ 745 reqsk_fastopen_remove(sk, req, true); 746 tcp_reset(sk); 747 } 748 if (!fastopen) { 749 inet_csk_reqsk_queue_drop(sk, req, prev); 750 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS); 751 } 752 return NULL; 753 } 754 EXPORT_SYMBOL(tcp_check_req); 755 756 /* 757 * Queue segment on the new socket if the new socket is active, 758 * otherwise we just shortcircuit this and continue with 759 * the new socket. 760 * 761 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV 762 * when entering. But other states are possible due to a race condition 763 * where after __inet_lookup_established() fails but before the listener 764 * locked is obtained, other packets cause the same connection to 765 * be created. 766 */ 767 768 int tcp_child_process(struct sock *parent, struct sock *child, 769 struct sk_buff *skb) 770 { 771 int ret = 0; 772 int state = child->sk_state; 773 774 if (!sock_owned_by_user(child)) { 775 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb), 776 skb->len); 777 /* Wakeup parent, send SIGIO */ 778 if (state == TCP_SYN_RECV && child->sk_state != state) 779 parent->sk_data_ready(parent); 780 } else { 781 /* Alas, it is possible again, because we do lookup 782 * in main socket hash table and lock on listening 783 * socket does not protect us more. 784 */ 785 __sk_add_backlog(child, skb); 786 } 787 788 bh_unlock_sock(child); 789 sock_put(child); 790 return ret; 791 } 792 EXPORT_SYMBOL(tcp_child_process); 793