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
tcp_in_window(u32 seq,u32 end_seq,u32 s_win,u32 e_win)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
tcp_timewait_state_process(struct inet_timewait_sock * tw,struct sk_buff * skb,const struct tcphdr * th)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 */
tcp_time_wait(struct sock * sk,int state,int timeo)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 struct inet6_timewait_sock *tw6;
297
298 tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
299 tw6 = inet6_twsk((struct sock *)tw);
300 tw6->tw_v6_daddr = np->daddr;
301 tw6->tw_v6_rcv_saddr = np->rcv_saddr;
302 tw->tw_tclass = np->tclass;
303 tw->tw_ipv6only = np->ipv6only;
304 }
305 #endif
306
307 #ifdef CONFIG_TCP_MD5SIG
308 /*
309 * The timewait bucket does not have the key DB from the
310 * sock structure. We just make a quick copy of the
311 * md5 key being used (if indeed we are using one)
312 * so the timewait ack generating code has the key.
313 */
314 do {
315 struct tcp_md5sig_key *key;
316 tcptw->tw_md5_key = NULL;
317 key = tp->af_specific->md5_lookup(sk, sk);
318 if (key != NULL) {
319 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
320 if (tcptw->tw_md5_key && tcp_alloc_md5sig_pool(sk) == NULL)
321 BUG();
322 }
323 } while (0);
324 #endif
325
326 /* Linkage updates. */
327 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
328
329 /* Get the TIME_WAIT timeout firing. */
330 if (timeo < rto)
331 timeo = rto;
332
333 if (recycle_ok) {
334 tw->tw_timeout = rto;
335 } else {
336 tw->tw_timeout = TCP_TIMEWAIT_LEN;
337 if (state == TCP_TIME_WAIT)
338 timeo = TCP_TIMEWAIT_LEN;
339 }
340
341 inet_twsk_schedule(tw, &tcp_death_row, timeo,
342 TCP_TIMEWAIT_LEN);
343 inet_twsk_put(tw);
344 } else {
345 /* Sorry, if we're out of memory, just CLOSE this
346 * socket up. We've got bigger problems than
347 * non-graceful socket closings.
348 */
349 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
350 }
351
352 tcp_update_metrics(sk);
353 tcp_done(sk);
354 }
355
tcp_twsk_destructor(struct sock * sk)356 void tcp_twsk_destructor(struct sock *sk)
357 {
358 #ifdef CONFIG_TCP_MD5SIG
359 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
360
361 if (twsk->tw_md5_key) {
362 tcp_free_md5sig_pool();
363 kfree_rcu(twsk->tw_md5_key, rcu);
364 }
365 #endif
366 }
367 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
368
TCP_ECN_openreq_child(struct tcp_sock * tp,struct request_sock * req)369 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
370 struct request_sock *req)
371 {
372 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
373 }
374
375 /* This is not only more efficient than what we used to do, it eliminates
376 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
377 *
378 * Actually, we could lots of memory writes here. tp of listening
379 * socket contains all necessary default parameters.
380 */
tcp_create_openreq_child(struct sock * sk,struct request_sock * req,struct sk_buff * skb)381 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
382 {
383 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
384
385 if (newsk != NULL) {
386 const struct inet_request_sock *ireq = inet_rsk(req);
387 struct tcp_request_sock *treq = tcp_rsk(req);
388 struct inet_connection_sock *newicsk = inet_csk(newsk);
389 struct tcp_sock *newtp = tcp_sk(newsk);
390
391 /* Now setup tcp_sock */
392 newtp->pred_flags = 0;
393
394 newtp->rcv_wup = newtp->copied_seq =
395 newtp->rcv_nxt = treq->rcv_isn + 1;
396
397 newtp->snd_sml = newtp->snd_una =
398 newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
399
400 tcp_prequeue_init(newtp);
401 INIT_LIST_HEAD(&newtp->tsq_node);
402
403 tcp_init_wl(newtp, treq->rcv_isn);
404
405 newtp->srtt = 0;
406 newtp->mdev = TCP_TIMEOUT_INIT;
407 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
408
409 newtp->packets_out = 0;
410 newtp->retrans_out = 0;
411 newtp->sacked_out = 0;
412 newtp->fackets_out = 0;
413 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
414 tcp_enable_early_retrans(newtp);
415 newtp->tlp_high_seq = 0;
416
417 /* So many TCP implementations out there (incorrectly) count the
418 * initial SYN frame in their delayed-ACK and congestion control
419 * algorithms that we must have the following bandaid to talk
420 * efficiently to them. -DaveM
421 */
422 newtp->snd_cwnd = TCP_INIT_CWND;
423 newtp->snd_cwnd_cnt = 0;
424
425 if (newicsk->icsk_ca_ops != &tcp_init_congestion_ops &&
426 !try_module_get(newicsk->icsk_ca_ops->owner))
427 newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
428
429 tcp_set_ca_state(newsk, TCP_CA_Open);
430 tcp_init_xmit_timers(newsk);
431 skb_queue_head_init(&newtp->out_of_order_queue);
432 newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
433
434 newtp->rx_opt.saw_tstamp = 0;
435
436 newtp->rx_opt.dsack = 0;
437 newtp->rx_opt.num_sacks = 0;
438
439 newtp->urg_data = 0;
440
441 if (sock_flag(newsk, SOCK_KEEPOPEN))
442 inet_csk_reset_keepalive_timer(newsk,
443 keepalive_time_when(newtp));
444
445 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
446 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
447 if (sysctl_tcp_fack)
448 tcp_enable_fack(newtp);
449 }
450 newtp->window_clamp = req->window_clamp;
451 newtp->rcv_ssthresh = req->rcv_wnd;
452 newtp->rcv_wnd = req->rcv_wnd;
453 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
454 if (newtp->rx_opt.wscale_ok) {
455 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
456 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
457 } else {
458 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
459 newtp->window_clamp = min(newtp->window_clamp, 65535U);
460 }
461 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
462 newtp->rx_opt.snd_wscale);
463 newtp->max_window = newtp->snd_wnd;
464
465 if (newtp->rx_opt.tstamp_ok) {
466 newtp->rx_opt.ts_recent = req->ts_recent;
467 newtp->rx_opt.ts_recent_stamp = get_seconds();
468 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
469 } else {
470 newtp->rx_opt.ts_recent_stamp = 0;
471 newtp->tcp_header_len = sizeof(struct tcphdr);
472 }
473 newtp->tsoffset = 0;
474 #ifdef CONFIG_TCP_MD5SIG
475 newtp->md5sig_info = NULL; /*XXX*/
476 if (newtp->af_specific->md5_lookup(sk, newsk))
477 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
478 #endif
479 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
480 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
481 newtp->rx_opt.mss_clamp = req->mss;
482 TCP_ECN_openreq_child(newtp, req);
483 newtp->fastopen_rsk = NULL;
484 newtp->syn_data_acked = 0;
485
486 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
487 }
488 return newsk;
489 }
490 EXPORT_SYMBOL(tcp_create_openreq_child);
491
492 /*
493 * Process an incoming packet for SYN_RECV sockets represented as a
494 * request_sock. Normally sk is the listener socket but for TFO it
495 * points to the child socket.
496 *
497 * XXX (TFO) - The current impl contains a special check for ack
498 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
499 *
500 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
501 */
502
tcp_check_req(struct sock * sk,struct sk_buff * skb,struct request_sock * req,struct request_sock ** prev,bool fastopen)503 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
504 struct request_sock *req,
505 struct request_sock **prev,
506 bool fastopen)
507 {
508 struct tcp_options_received tmp_opt;
509 struct sock *child;
510 const struct tcphdr *th = tcp_hdr(skb);
511 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
512 bool paws_reject = false;
513
514 BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
515
516 tmp_opt.saw_tstamp = 0;
517 if (th->doff > (sizeof(struct tcphdr)>>2)) {
518 tcp_parse_options(skb, &tmp_opt, 0, NULL);
519
520 if (tmp_opt.saw_tstamp) {
521 tmp_opt.ts_recent = req->ts_recent;
522 /* We do not store true stamp, but it is not required,
523 * it can be estimated (approximately)
524 * from another data.
525 */
526 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
527 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
528 }
529 }
530
531 /* Check for pure retransmitted SYN. */
532 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
533 flg == TCP_FLAG_SYN &&
534 !paws_reject) {
535 /*
536 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
537 * this case on figure 6 and figure 8, but formal
538 * protocol description says NOTHING.
539 * To be more exact, it says that we should send ACK,
540 * because this segment (at least, if it has no data)
541 * is out of window.
542 *
543 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
544 * describe SYN-RECV state. All the description
545 * is wrong, we cannot believe to it and should
546 * rely only on common sense and implementation
547 * experience.
548 *
549 * Enforce "SYN-ACK" according to figure 8, figure 6
550 * of RFC793, fixed by RFC1122.
551 *
552 * Note that even if there is new data in the SYN packet
553 * they will be thrown away too.
554 *
555 * Reset timer after retransmitting SYNACK, similar to
556 * the idea of fast retransmit in recovery.
557 */
558 if (!inet_rtx_syn_ack(sk, req))
559 req->expires = min(TCP_TIMEOUT_INIT << req->num_timeout,
560 TCP_RTO_MAX) + jiffies;
561 return NULL;
562 }
563
564 /* Further reproduces section "SEGMENT ARRIVES"
565 for state SYN-RECEIVED of RFC793.
566 It is broken, however, it does not work only
567 when SYNs are crossed.
568
569 You would think that SYN crossing is impossible here, since
570 we should have a SYN_SENT socket (from connect()) on our end,
571 but this is not true if the crossed SYNs were sent to both
572 ends by a malicious third party. We must defend against this,
573 and to do that we first verify the ACK (as per RFC793, page
574 36) and reset if it is invalid. Is this a true full defense?
575 To convince ourselves, let us consider a way in which the ACK
576 test can still pass in this 'malicious crossed SYNs' case.
577 Malicious sender sends identical SYNs (and thus identical sequence
578 numbers) to both A and B:
579
580 A: gets SYN, seq=7
581 B: gets SYN, seq=7
582
583 By our good fortune, both A and B select the same initial
584 send sequence number of seven :-)
585
586 A: sends SYN|ACK, seq=7, ack_seq=8
587 B: sends SYN|ACK, seq=7, ack_seq=8
588
589 So we are now A eating this SYN|ACK, ACK test passes. So
590 does sequence test, SYN is truncated, and thus we consider
591 it a bare ACK.
592
593 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
594 bare ACK. Otherwise, we create an established connection. Both
595 ends (listening sockets) accept the new incoming connection and try
596 to talk to each other. 8-)
597
598 Note: This case is both harmless, and rare. Possibility is about the
599 same as us discovering intelligent life on another plant tomorrow.
600
601 But generally, we should (RFC lies!) to accept ACK
602 from SYNACK both here and in tcp_rcv_state_process().
603 tcp_rcv_state_process() does not, hence, we do not too.
604
605 Note that the case is absolutely generic:
606 we cannot optimize anything here without
607 violating protocol. All the checks must be made
608 before attempt to create socket.
609 */
610
611 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
612 * and the incoming segment acknowledges something not yet
613 * sent (the segment carries an unacceptable ACK) ...
614 * a reset is sent."
615 *
616 * Invalid ACK: reset will be sent by listening socket.
617 * Note that the ACK validity check for a Fast Open socket is done
618 * elsewhere and is checked directly against the child socket rather
619 * than req because user data may have been sent out.
620 */
621 if ((flg & TCP_FLAG_ACK) && !fastopen &&
622 (TCP_SKB_CB(skb)->ack_seq !=
623 tcp_rsk(req)->snt_isn + 1))
624 return sk;
625
626 /* Also, it would be not so bad idea to check rcv_tsecr, which
627 * is essentially ACK extension and too early or too late values
628 * should cause reset in unsynchronized states.
629 */
630
631 /* RFC793: "first check sequence number". */
632
633 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
634 tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
635 /* Out of window: send ACK and drop. */
636 if (!(flg & TCP_FLAG_RST))
637 req->rsk_ops->send_ack(sk, skb, req);
638 if (paws_reject)
639 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
640 return NULL;
641 }
642
643 /* In sequence, PAWS is OK. */
644
645 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
646 req->ts_recent = tmp_opt.rcv_tsval;
647
648 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
649 /* Truncate SYN, it is out of window starting
650 at tcp_rsk(req)->rcv_isn + 1. */
651 flg &= ~TCP_FLAG_SYN;
652 }
653
654 /* RFC793: "second check the RST bit" and
655 * "fourth, check the SYN bit"
656 */
657 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
658 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
659 goto embryonic_reset;
660 }
661
662 /* ACK sequence verified above, just make sure ACK is
663 * set. If ACK not set, just silently drop the packet.
664 *
665 * XXX (TFO) - if we ever allow "data after SYN", the
666 * following check needs to be removed.
667 */
668 if (!(flg & TCP_FLAG_ACK))
669 return NULL;
670
671 /* Got ACK for our SYNACK, so update baseline for SYNACK RTT sample. */
672 if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
673 tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
674 else if (req->num_retrans) /* don't take RTT sample if retrans && ~TS */
675 tcp_rsk(req)->snt_synack = 0;
676
677 /* For Fast Open no more processing is needed (sk is the
678 * child socket).
679 */
680 if (fastopen)
681 return sk;
682
683 /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
684 if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
685 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
686 inet_rsk(req)->acked = 1;
687 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
688 return NULL;
689 }
690
691 /* OK, ACK is valid, create big socket and
692 * feed this segment to it. It will repeat all
693 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
694 * ESTABLISHED STATE. If it will be dropped after
695 * socket is created, wait for troubles.
696 */
697 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
698 if (child == NULL)
699 goto listen_overflow;
700
701 inet_csk_reqsk_queue_unlink(sk, req, prev);
702 inet_csk_reqsk_queue_removed(sk, req);
703
704 inet_csk_reqsk_queue_add(sk, req, child);
705 return child;
706
707 listen_overflow:
708 if (!sysctl_tcp_abort_on_overflow) {
709 inet_rsk(req)->acked = 1;
710 return NULL;
711 }
712
713 embryonic_reset:
714 if (!(flg & TCP_FLAG_RST)) {
715 /* Received a bad SYN pkt - for TFO We try not to reset
716 * the local connection unless it's really necessary to
717 * avoid becoming vulnerable to outside attack aiming at
718 * resetting legit local connections.
719 */
720 req->rsk_ops->send_reset(sk, skb);
721 } else if (fastopen) { /* received a valid RST pkt */
722 reqsk_fastopen_remove(sk, req, true);
723 tcp_reset(sk);
724 }
725 if (!fastopen) {
726 inet_csk_reqsk_queue_drop(sk, req, prev);
727 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
728 }
729 return NULL;
730 }
731 EXPORT_SYMBOL(tcp_check_req);
732
733 /*
734 * Queue segment on the new socket if the new socket is active,
735 * otherwise we just shortcircuit this and continue with
736 * the new socket.
737 *
738 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
739 * when entering. But other states are possible due to a race condition
740 * where after __inet_lookup_established() fails but before the listener
741 * locked is obtained, other packets cause the same connection to
742 * be created.
743 */
744
tcp_child_process(struct sock * parent,struct sock * child,struct sk_buff * skb)745 int tcp_child_process(struct sock *parent, struct sock *child,
746 struct sk_buff *skb)
747 {
748 int ret = 0;
749 int state = child->sk_state;
750
751 if (!sock_owned_by_user(child)) {
752 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
753 skb->len);
754 /* Wakeup parent, send SIGIO */
755 if (state == TCP_SYN_RECV && child->sk_state != state)
756 parent->sk_data_ready(parent, 0);
757 } else {
758 /* Alas, it is possible again, because we do lookup
759 * in main socket hash table and lock on listening
760 * socket does not protect us more.
761 */
762 __sk_add_backlog(child, skb);
763 }
764
765 bh_unlock_sock(child);
766 sock_put(child);
767 return ret;
768 }
769 EXPORT_SYMBOL(tcp_child_process);
770