xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision e3b9f1e8)
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 <linux/static_key.h>
27 #include <net/tcp.h>
28 #include <net/inet_common.h>
29 #include <net/xfrm.h>
30 #include <net/busy_poll.h>
31 
32 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
33 {
34 	if (seq == s_win)
35 		return true;
36 	if (after(end_seq, s_win) && before(seq, e_win))
37 		return true;
38 	return seq == e_win && seq == end_seq;
39 }
40 
41 static enum tcp_tw_status
42 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
43 				  const struct sk_buff *skb, int mib_idx)
44 {
45 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
46 
47 	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
48 				  &tcptw->tw_last_oow_ack_time)) {
49 		/* Send ACK. Note, we do not put the bucket,
50 		 * it will be released by caller.
51 		 */
52 		return TCP_TW_ACK;
53 	}
54 
55 	/* We are rate-limiting, so just release the tw sock and drop skb. */
56 	inet_twsk_put(tw);
57 	return TCP_TW_SUCCESS;
58 }
59 
60 /*
61  * * Main purpose of TIME-WAIT state is to close connection gracefully,
62  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
63  *   (and, probably, tail of data) and one or more our ACKs are lost.
64  * * What is TIME-WAIT timeout? It is associated with maximal packet
65  *   lifetime in the internet, which results in wrong conclusion, that
66  *   it is set to catch "old duplicate segments" wandering out of their path.
67  *   It is not quite correct. This timeout is calculated so that it exceeds
68  *   maximal retransmission timeout enough to allow to lose one (or more)
69  *   segments sent by peer and our ACKs. This time may be calculated from RTO.
70  * * When TIME-WAIT socket receives RST, it means that another end
71  *   finally closed and we are allowed to kill TIME-WAIT too.
72  * * Second purpose of TIME-WAIT is catching old duplicate segments.
73  *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
74  *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
75  * * If we invented some more clever way to catch duplicates
76  *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
77  *
78  * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
79  * When you compare it to RFCs, please, read section SEGMENT ARRIVES
80  * from the very beginning.
81  *
82  * NOTE. With recycling (and later with fin-wait-2) TW bucket
83  * is _not_ stateless. It means, that strictly speaking we must
84  * spinlock it. I do not want! Well, probability of misbehaviour
85  * is ridiculously low and, seems, we could use some mb() tricks
86  * to avoid misread sequence numbers, states etc.  --ANK
87  *
88  * We don't need to initialize tmp_out.sack_ok as we don't use the results
89  */
90 enum tcp_tw_status
91 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
92 			   const struct tcphdr *th)
93 {
94 	struct tcp_options_received tmp_opt;
95 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
96 	bool paws_reject = false;
97 
98 	tmp_opt.saw_tstamp = 0;
99 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
100 		tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
101 
102 		if (tmp_opt.saw_tstamp) {
103 			if (tmp_opt.rcv_tsecr)
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_timewait_check_oow_rate_limit(
120 				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
121 
122 		if (th->rst)
123 			goto kill;
124 
125 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
126 			return TCP_TW_RST;
127 
128 		/* Dup ACK? */
129 		if (!th->ack ||
130 		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
131 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
132 			inet_twsk_put(tw);
133 			return TCP_TW_SUCCESS;
134 		}
135 
136 		/* New data or FIN. If new data arrive after half-duplex close,
137 		 * reset.
138 		 */
139 		if (!th->fin ||
140 		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
141 			return TCP_TW_RST;
142 
143 		/* FIN arrived, enter true time-wait state. */
144 		tw->tw_substate	  = TCP_TIME_WAIT;
145 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
146 		if (tmp_opt.saw_tstamp) {
147 			tcptw->tw_ts_recent_stamp = get_seconds();
148 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
149 		}
150 
151 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
152 		return TCP_TW_ACK;
153 	}
154 
155 	/*
156 	 *	Now real TIME-WAIT state.
157 	 *
158 	 *	RFC 1122:
159 	 *	"When a connection is [...] on TIME-WAIT state [...]
160 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
161 	 *	reopen the connection directly, if it:
162 	 *
163 	 *	(1)  assigns its initial sequence number for the new
164 	 *	connection to be larger than the largest sequence
165 	 *	number it used on the previous connection incarnation,
166 	 *	and
167 	 *
168 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
169 	 *	to be an old duplicate".
170 	 */
171 
172 	if (!paws_reject &&
173 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
174 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
175 		/* In window segment, it may be only reset or bare ack. */
176 
177 		if (th->rst) {
178 			/* This is TIME_WAIT assassination, in two flavors.
179 			 * Oh well... nobody has a sufficient solution to this
180 			 * protocol bug yet.
181 			 */
182 			if (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) {
183 kill:
184 				inet_twsk_deschedule_put(tw);
185 				return TCP_TW_SUCCESS;
186 			}
187 		}
188 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
189 
190 		if (tmp_opt.saw_tstamp) {
191 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
192 			tcptw->tw_ts_recent_stamp = get_seconds();
193 		}
194 
195 		inet_twsk_put(tw);
196 		return TCP_TW_SUCCESS;
197 	}
198 
199 	/* Out of window segment.
200 
201 	   All the segments are ACKed immediately.
202 
203 	   The only exception is new SYN. We accept it, if it is
204 	   not old duplicate and we are not in danger to be killed
205 	   by delayed old duplicates. RFC check is that it has
206 	   newer sequence number works at rates <40Mbit/sec.
207 	   However, if paws works, it is reliable AND even more,
208 	   we even may relax silly seq space cutoff.
209 
210 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
211 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
212 	   we must return socket to time-wait state. It is not good,
213 	   but not fatal yet.
214 	 */
215 
216 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
217 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
218 	     (tmp_opt.saw_tstamp &&
219 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
220 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
221 		if (isn == 0)
222 			isn++;
223 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
224 		return TCP_TW_SYN;
225 	}
226 
227 	if (paws_reject)
228 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
229 
230 	if (!th->rst) {
231 		/* In this case we must reset the TIMEWAIT timer.
232 		 *
233 		 * If it is ACKless SYN it may be both old duplicate
234 		 * and new good SYN with random sequence number <rcv_nxt.
235 		 * Do not reschedule in the last case.
236 		 */
237 		if (paws_reject || th->ack)
238 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
239 
240 		return tcp_timewait_check_oow_rate_limit(
241 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
242 	}
243 	inet_twsk_put(tw);
244 	return TCP_TW_SUCCESS;
245 }
246 EXPORT_SYMBOL(tcp_timewait_state_process);
247 
248 /*
249  * Move a socket to time-wait or dead fin-wait-2 state.
250  */
251 void tcp_time_wait(struct sock *sk, int state, int timeo)
252 {
253 	const struct inet_connection_sock *icsk = inet_csk(sk);
254 	const struct tcp_sock *tp = tcp_sk(sk);
255 	struct inet_timewait_sock *tw;
256 	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
257 
258 	tw = inet_twsk_alloc(sk, tcp_death_row, state);
259 
260 	if (tw) {
261 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
262 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
263 		struct inet_sock *inet = inet_sk(sk);
264 
265 		tw->tw_transparent	= inet->transparent;
266 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
267 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
268 		tcptw->tw_snd_nxt	= tp->snd_nxt;
269 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
270 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
271 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
272 		tcptw->tw_ts_offset	= tp->tsoffset;
273 		tcptw->tw_last_oow_ack_time = 0;
274 
275 #if IS_ENABLED(CONFIG_IPV6)
276 		if (tw->tw_family == PF_INET6) {
277 			struct ipv6_pinfo *np = inet6_sk(sk);
278 
279 			tw->tw_v6_daddr = sk->sk_v6_daddr;
280 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
281 			tw->tw_tclass = np->tclass;
282 			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
283 			tw->tw_ipv6only = sk->sk_ipv6only;
284 		}
285 #endif
286 
287 #ifdef CONFIG_TCP_MD5SIG
288 		/*
289 		 * The timewait bucket does not have the key DB from the
290 		 * sock structure. We just make a quick copy of the
291 		 * md5 key being used (if indeed we are using one)
292 		 * so the timewait ack generating code has the key.
293 		 */
294 		do {
295 			struct tcp_md5sig_key *key;
296 			tcptw->tw_md5_key = NULL;
297 			key = tp->af_specific->md5_lookup(sk, sk);
298 			if (key) {
299 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
300 				BUG_ON(tcptw->tw_md5_key && !tcp_alloc_md5sig_pool());
301 			}
302 		} while (0);
303 #endif
304 
305 		/* Get the TIME_WAIT timeout firing. */
306 		if (timeo < rto)
307 			timeo = rto;
308 
309 		tw->tw_timeout = TCP_TIMEWAIT_LEN;
310 		if (state == TCP_TIME_WAIT)
311 			timeo = TCP_TIMEWAIT_LEN;
312 
313 		/* tw_timer is pinned, so we need to make sure BH are disabled
314 		 * in following section, otherwise timer handler could run before
315 		 * we complete the initialization.
316 		 */
317 		local_bh_disable();
318 		inet_twsk_schedule(tw, timeo);
319 		/* Linkage updates.
320 		 * Note that access to tw after this point is illegal.
321 		 */
322 		inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
323 		local_bh_enable();
324 	} else {
325 		/* Sorry, if we're out of memory, just CLOSE this
326 		 * socket up.  We've got bigger problems than
327 		 * non-graceful socket closings.
328 		 */
329 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
330 	}
331 
332 	tcp_update_metrics(sk);
333 	tcp_done(sk);
334 }
335 
336 void tcp_twsk_destructor(struct sock *sk)
337 {
338 #ifdef CONFIG_TCP_MD5SIG
339 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
340 
341 	if (twsk->tw_md5_key)
342 		kfree_rcu(twsk->tw_md5_key, rcu);
343 #endif
344 }
345 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
346 
347 /* Warning : This function is called without sk_listener being locked.
348  * Be sure to read socket fields once, as their value could change under us.
349  */
350 void tcp_openreq_init_rwin(struct request_sock *req,
351 			   const struct sock *sk_listener,
352 			   const struct dst_entry *dst)
353 {
354 	struct inet_request_sock *ireq = inet_rsk(req);
355 	const struct tcp_sock *tp = tcp_sk(sk_listener);
356 	int full_space = tcp_full_space(sk_listener);
357 	u32 window_clamp;
358 	__u8 rcv_wscale;
359 	u32 rcv_wnd;
360 	int mss;
361 
362 	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
363 	window_clamp = READ_ONCE(tp->window_clamp);
364 	/* Set this up on the first call only */
365 	req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
366 
367 	/* limit the window selection if the user enforce a smaller rx buffer */
368 	if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
369 	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
370 		req->rsk_window_clamp = full_space;
371 
372 	rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
373 	if (rcv_wnd == 0)
374 		rcv_wnd = dst_metric(dst, RTAX_INITRWND);
375 	else if (full_space < rcv_wnd * mss)
376 		full_space = rcv_wnd * mss;
377 
378 	/* tcp_full_space because it is guaranteed to be the first packet */
379 	tcp_select_initial_window(sk_listener, full_space,
380 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
381 		&req->rsk_rcv_wnd,
382 		&req->rsk_window_clamp,
383 		ireq->wscale_ok,
384 		&rcv_wscale,
385 		rcv_wnd);
386 	ireq->rcv_wscale = rcv_wscale;
387 }
388 EXPORT_SYMBOL(tcp_openreq_init_rwin);
389 
390 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
391 				  const struct request_sock *req)
392 {
393 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
394 }
395 
396 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
397 {
398 	struct inet_connection_sock *icsk = inet_csk(sk);
399 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
400 	bool ca_got_dst = false;
401 
402 	if (ca_key != TCP_CA_UNSPEC) {
403 		const struct tcp_congestion_ops *ca;
404 
405 		rcu_read_lock();
406 		ca = tcp_ca_find_key(ca_key);
407 		if (likely(ca && try_module_get(ca->owner))) {
408 			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
409 			icsk->icsk_ca_ops = ca;
410 			ca_got_dst = true;
411 		}
412 		rcu_read_unlock();
413 	}
414 
415 	/* If no valid choice made yet, assign current system default ca. */
416 	if (!ca_got_dst &&
417 	    (!icsk->icsk_ca_setsockopt ||
418 	     !try_module_get(icsk->icsk_ca_ops->owner)))
419 		tcp_assign_congestion_control(sk);
420 
421 	tcp_set_ca_state(sk, TCP_CA_Open);
422 }
423 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
424 
425 static void smc_check_reset_syn_req(struct tcp_sock *oldtp,
426 				    struct request_sock *req,
427 				    struct tcp_sock *newtp)
428 {
429 #if IS_ENABLED(CONFIG_SMC)
430 	struct inet_request_sock *ireq;
431 
432 	if (static_branch_unlikely(&tcp_have_smc)) {
433 		ireq = inet_rsk(req);
434 		if (oldtp->syn_smc && !ireq->smc_ok)
435 			newtp->syn_smc = 0;
436 	}
437 #endif
438 }
439 
440 /* This is not only more efficient than what we used to do, it eliminates
441  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
442  *
443  * Actually, we could lots of memory writes here. tp of listening
444  * socket contains all necessary default parameters.
445  */
446 struct sock *tcp_create_openreq_child(const struct sock *sk,
447 				      struct request_sock *req,
448 				      struct sk_buff *skb)
449 {
450 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
451 
452 	if (newsk) {
453 		const struct inet_request_sock *ireq = inet_rsk(req);
454 		struct tcp_request_sock *treq = tcp_rsk(req);
455 		struct inet_connection_sock *newicsk = inet_csk(newsk);
456 		struct tcp_sock *newtp = tcp_sk(newsk);
457 		struct tcp_sock *oldtp = tcp_sk(sk);
458 
459 		smc_check_reset_syn_req(oldtp, req, newtp);
460 
461 		/* Now setup tcp_sock */
462 		newtp->pred_flags = 0;
463 
464 		newtp->rcv_wup = newtp->copied_seq =
465 		newtp->rcv_nxt = treq->rcv_isn + 1;
466 		newtp->segs_in = 1;
467 
468 		newtp->snd_sml = newtp->snd_una =
469 		newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
470 
471 		INIT_LIST_HEAD(&newtp->tsq_node);
472 		INIT_LIST_HEAD(&newtp->tsorted_sent_queue);
473 
474 		tcp_init_wl(newtp, treq->rcv_isn);
475 
476 		newtp->srtt_us = 0;
477 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
478 		minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
479 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
480 		newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
481 
482 		newtp->packets_out = 0;
483 		newtp->retrans_out = 0;
484 		newtp->sacked_out = 0;
485 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
486 		newtp->tlp_high_seq = 0;
487 		newtp->lsndtime = tcp_jiffies32;
488 		newsk->sk_txhash = treq->txhash;
489 		newtp->last_oow_ack_time = 0;
490 		newtp->total_retrans = req->num_retrans;
491 
492 		/* So many TCP implementations out there (incorrectly) count the
493 		 * initial SYN frame in their delayed-ACK and congestion control
494 		 * algorithms that we must have the following bandaid to talk
495 		 * efficiently to them.  -DaveM
496 		 */
497 		newtp->snd_cwnd = TCP_INIT_CWND;
498 		newtp->snd_cwnd_cnt = 0;
499 
500 		/* There's a bubble in the pipe until at least the first ACK. */
501 		newtp->app_limited = ~0U;
502 
503 		tcp_init_xmit_timers(newsk);
504 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
505 
506 		newtp->rx_opt.saw_tstamp = 0;
507 
508 		newtp->rx_opt.dsack = 0;
509 		newtp->rx_opt.num_sacks = 0;
510 
511 		newtp->urg_data = 0;
512 
513 		if (sock_flag(newsk, SOCK_KEEPOPEN))
514 			inet_csk_reset_keepalive_timer(newsk,
515 						       keepalive_time_when(newtp));
516 
517 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
518 		newtp->rx_opt.sack_ok = ireq->sack_ok;
519 		newtp->window_clamp = req->rsk_window_clamp;
520 		newtp->rcv_ssthresh = req->rsk_rcv_wnd;
521 		newtp->rcv_wnd = req->rsk_rcv_wnd;
522 		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
523 		if (newtp->rx_opt.wscale_ok) {
524 			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
525 			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
526 		} else {
527 			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
528 			newtp->window_clamp = min(newtp->window_clamp, 65535U);
529 		}
530 		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
531 				  newtp->rx_opt.snd_wscale);
532 		newtp->max_window = newtp->snd_wnd;
533 
534 		if (newtp->rx_opt.tstamp_ok) {
535 			newtp->rx_opt.ts_recent = req->ts_recent;
536 			newtp->rx_opt.ts_recent_stamp = get_seconds();
537 			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
538 		} else {
539 			newtp->rx_opt.ts_recent_stamp = 0;
540 			newtp->tcp_header_len = sizeof(struct tcphdr);
541 		}
542 		newtp->tsoffset = treq->ts_off;
543 #ifdef CONFIG_TCP_MD5SIG
544 		newtp->md5sig_info = NULL;	/*XXX*/
545 		if (newtp->af_specific->md5_lookup(sk, newsk))
546 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
547 #endif
548 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
549 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
550 		newtp->rx_opt.mss_clamp = req->mss;
551 		tcp_ecn_openreq_child(newtp, req);
552 		newtp->fastopen_req = NULL;
553 		newtp->fastopen_rsk = NULL;
554 		newtp->syn_data_acked = 0;
555 		newtp->rack.mstamp = 0;
556 		newtp->rack.advanced = 0;
557 		newtp->rack.reo_wnd_steps = 1;
558 		newtp->rack.last_delivered = 0;
559 		newtp->rack.reo_wnd_persist = 0;
560 		newtp->rack.dsack_seen = 0;
561 
562 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
563 	}
564 	return newsk;
565 }
566 EXPORT_SYMBOL(tcp_create_openreq_child);
567 
568 /*
569  * Process an incoming packet for SYN_RECV sockets represented as a
570  * request_sock. Normally sk is the listener socket but for TFO it
571  * points to the child socket.
572  *
573  * XXX (TFO) - The current impl contains a special check for ack
574  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
575  *
576  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
577  */
578 
579 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
580 			   struct request_sock *req,
581 			   bool fastopen)
582 {
583 	struct tcp_options_received tmp_opt;
584 	struct sock *child;
585 	const struct tcphdr *th = tcp_hdr(skb);
586 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
587 	bool paws_reject = false;
588 	bool own_req;
589 
590 	tmp_opt.saw_tstamp = 0;
591 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
592 		tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
593 
594 		if (tmp_opt.saw_tstamp) {
595 			tmp_opt.ts_recent = req->ts_recent;
596 			if (tmp_opt.rcv_tsecr)
597 				tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
598 			/* We do not store true stamp, but it is not required,
599 			 * it can be estimated (approximately)
600 			 * from another data.
601 			 */
602 			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
603 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
604 		}
605 	}
606 
607 	/* Check for pure retransmitted SYN. */
608 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
609 	    flg == TCP_FLAG_SYN &&
610 	    !paws_reject) {
611 		/*
612 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
613 		 * this case on figure 6 and figure 8, but formal
614 		 * protocol description says NOTHING.
615 		 * To be more exact, it says that we should send ACK,
616 		 * because this segment (at least, if it has no data)
617 		 * is out of window.
618 		 *
619 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
620 		 *  describe SYN-RECV state. All the description
621 		 *  is wrong, we cannot believe to it and should
622 		 *  rely only on common sense and implementation
623 		 *  experience.
624 		 *
625 		 * Enforce "SYN-ACK" according to figure 8, figure 6
626 		 * of RFC793, fixed by RFC1122.
627 		 *
628 		 * Note that even if there is new data in the SYN packet
629 		 * they will be thrown away too.
630 		 *
631 		 * Reset timer after retransmitting SYNACK, similar to
632 		 * the idea of fast retransmit in recovery.
633 		 */
634 		if (!tcp_oow_rate_limited(sock_net(sk), skb,
635 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
636 					  &tcp_rsk(req)->last_oow_ack_time) &&
637 
638 		    !inet_rtx_syn_ack(sk, req)) {
639 			unsigned long expires = jiffies;
640 
641 			expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
642 				       TCP_RTO_MAX);
643 			if (!fastopen)
644 				mod_timer_pending(&req->rsk_timer, expires);
645 			else
646 				req->rsk_timer.expires = expires;
647 		}
648 		return NULL;
649 	}
650 
651 	/* Further reproduces section "SEGMENT ARRIVES"
652 	   for state SYN-RECEIVED of RFC793.
653 	   It is broken, however, it does not work only
654 	   when SYNs are crossed.
655 
656 	   You would think that SYN crossing is impossible here, since
657 	   we should have a SYN_SENT socket (from connect()) on our end,
658 	   but this is not true if the crossed SYNs were sent to both
659 	   ends by a malicious third party.  We must defend against this,
660 	   and to do that we first verify the ACK (as per RFC793, page
661 	   36) and reset if it is invalid.  Is this a true full defense?
662 	   To convince ourselves, let us consider a way in which the ACK
663 	   test can still pass in this 'malicious crossed SYNs' case.
664 	   Malicious sender sends identical SYNs (and thus identical sequence
665 	   numbers) to both A and B:
666 
667 		A: gets SYN, seq=7
668 		B: gets SYN, seq=7
669 
670 	   By our good fortune, both A and B select the same initial
671 	   send sequence number of seven :-)
672 
673 		A: sends SYN|ACK, seq=7, ack_seq=8
674 		B: sends SYN|ACK, seq=7, ack_seq=8
675 
676 	   So we are now A eating this SYN|ACK, ACK test passes.  So
677 	   does sequence test, SYN is truncated, and thus we consider
678 	   it a bare ACK.
679 
680 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
681 	   bare ACK.  Otherwise, we create an established connection.  Both
682 	   ends (listening sockets) accept the new incoming connection and try
683 	   to talk to each other. 8-)
684 
685 	   Note: This case is both harmless, and rare.  Possibility is about the
686 	   same as us discovering intelligent life on another plant tomorrow.
687 
688 	   But generally, we should (RFC lies!) to accept ACK
689 	   from SYNACK both here and in tcp_rcv_state_process().
690 	   tcp_rcv_state_process() does not, hence, we do not too.
691 
692 	   Note that the case is absolutely generic:
693 	   we cannot optimize anything here without
694 	   violating protocol. All the checks must be made
695 	   before attempt to create socket.
696 	 */
697 
698 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
699 	 *                  and the incoming segment acknowledges something not yet
700 	 *                  sent (the segment carries an unacceptable ACK) ...
701 	 *                  a reset is sent."
702 	 *
703 	 * Invalid ACK: reset will be sent by listening socket.
704 	 * Note that the ACK validity check for a Fast Open socket is done
705 	 * elsewhere and is checked directly against the child socket rather
706 	 * than req because user data may have been sent out.
707 	 */
708 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
709 	    (TCP_SKB_CB(skb)->ack_seq !=
710 	     tcp_rsk(req)->snt_isn + 1))
711 		return sk;
712 
713 	/* Also, it would be not so bad idea to check rcv_tsecr, which
714 	 * is essentially ACK extension and too early or too late values
715 	 * should cause reset in unsynchronized states.
716 	 */
717 
718 	/* RFC793: "first check sequence number". */
719 
720 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
721 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
722 		/* Out of window: send ACK and drop. */
723 		if (!(flg & TCP_FLAG_RST) &&
724 		    !tcp_oow_rate_limited(sock_net(sk), skb,
725 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
726 					  &tcp_rsk(req)->last_oow_ack_time))
727 			req->rsk_ops->send_ack(sk, skb, req);
728 		if (paws_reject)
729 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
730 		return NULL;
731 	}
732 
733 	/* In sequence, PAWS is OK. */
734 
735 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
736 		req->ts_recent = tmp_opt.rcv_tsval;
737 
738 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
739 		/* Truncate SYN, it is out of window starting
740 		   at tcp_rsk(req)->rcv_isn + 1. */
741 		flg &= ~TCP_FLAG_SYN;
742 	}
743 
744 	/* RFC793: "second check the RST bit" and
745 	 *	   "fourth, check the SYN bit"
746 	 */
747 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
748 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
749 		goto embryonic_reset;
750 	}
751 
752 	/* ACK sequence verified above, just make sure ACK is
753 	 * set.  If ACK not set, just silently drop the packet.
754 	 *
755 	 * XXX (TFO) - if we ever allow "data after SYN", the
756 	 * following check needs to be removed.
757 	 */
758 	if (!(flg & TCP_FLAG_ACK))
759 		return NULL;
760 
761 	/* For Fast Open no more processing is needed (sk is the
762 	 * child socket).
763 	 */
764 	if (fastopen)
765 		return sk;
766 
767 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
768 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
769 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
770 		inet_rsk(req)->acked = 1;
771 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
772 		return NULL;
773 	}
774 
775 	/* OK, ACK is valid, create big socket and
776 	 * feed this segment to it. It will repeat all
777 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
778 	 * ESTABLISHED STATE. If it will be dropped after
779 	 * socket is created, wait for troubles.
780 	 */
781 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
782 							 req, &own_req);
783 	if (!child)
784 		goto listen_overflow;
785 
786 	sock_rps_save_rxhash(child, skb);
787 	tcp_synack_rtt_meas(child, req);
788 	return inet_csk_complete_hashdance(sk, child, req, own_req);
789 
790 listen_overflow:
791 	if (!sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow) {
792 		inet_rsk(req)->acked = 1;
793 		return NULL;
794 	}
795 
796 embryonic_reset:
797 	if (!(flg & TCP_FLAG_RST)) {
798 		/* Received a bad SYN pkt - for TFO We try not to reset
799 		 * the local connection unless it's really necessary to
800 		 * avoid becoming vulnerable to outside attack aiming at
801 		 * resetting legit local connections.
802 		 */
803 		req->rsk_ops->send_reset(sk, skb);
804 	} else if (fastopen) { /* received a valid RST pkt */
805 		reqsk_fastopen_remove(sk, req, true);
806 		tcp_reset(sk);
807 	}
808 	if (!fastopen) {
809 		inet_csk_reqsk_queue_drop(sk, req);
810 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
811 	}
812 	return NULL;
813 }
814 EXPORT_SYMBOL(tcp_check_req);
815 
816 /*
817  * Queue segment on the new socket if the new socket is active,
818  * otherwise we just shortcircuit this and continue with
819  * the new socket.
820  *
821  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
822  * when entering. But other states are possible due to a race condition
823  * where after __inet_lookup_established() fails but before the listener
824  * locked is obtained, other packets cause the same connection to
825  * be created.
826  */
827 
828 int tcp_child_process(struct sock *parent, struct sock *child,
829 		      struct sk_buff *skb)
830 {
831 	int ret = 0;
832 	int state = child->sk_state;
833 
834 	/* record NAPI ID of child */
835 	sk_mark_napi_id(child, skb);
836 
837 	tcp_segs_in(tcp_sk(child), skb);
838 	if (!sock_owned_by_user(child)) {
839 		ret = tcp_rcv_state_process(child, skb);
840 		/* Wakeup parent, send SIGIO */
841 		if (state == TCP_SYN_RECV && child->sk_state != state)
842 			parent->sk_data_ready(parent);
843 	} else {
844 		/* Alas, it is possible again, because we do lookup
845 		 * in main socket hash table and lock on listening
846 		 * socket does not protect us more.
847 		 */
848 		__sk_add_backlog(child, skb);
849 	}
850 
851 	bh_unlock_sock(child);
852 	sock_put(child);
853 	return ret;
854 }
855 EXPORT_SYMBOL(tcp_child_process);
856