xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision a6ca5ac746d104019e76c29e69c2a1fc6dd2b29f)
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 #include <net/busy_poll.h>
30 
31 int sysctl_tcp_abort_on_overflow __read_mostly;
32 
33 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
34 {
35 	if (seq == s_win)
36 		return true;
37 	if (after(end_seq, s_win) && before(seq, e_win))
38 		return true;
39 	return seq == e_win && seq == end_seq;
40 }
41 
42 static enum tcp_tw_status
43 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
44 				  const struct sk_buff *skb, int mib_idx)
45 {
46 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
47 
48 	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
49 				  &tcptw->tw_last_oow_ack_time)) {
50 		/* Send ACK. Note, we do not put the bucket,
51 		 * it will be released by caller.
52 		 */
53 		return TCP_TW_ACK;
54 	}
55 
56 	/* We are rate-limiting, so just release the tw sock and drop skb. */
57 	inet_twsk_put(tw);
58 	return TCP_TW_SUCCESS;
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 			if (tmp_opt.rcv_tsecr)
105 				tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
106 			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
107 			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
108 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
109 		}
110 	}
111 
112 	if (tw->tw_substate == TCP_FIN_WAIT2) {
113 		/* Just repeat all the checks of tcp_rcv_state_process() */
114 
115 		/* Out of window, send ACK */
116 		if (paws_reject ||
117 		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
118 				   tcptw->tw_rcv_nxt,
119 				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
120 			return tcp_timewait_check_oow_rate_limit(
121 				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
122 
123 		if (th->rst)
124 			goto kill;
125 
126 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
127 			return TCP_TW_RST;
128 
129 		/* Dup ACK? */
130 		if (!th->ack ||
131 		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
132 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
133 			inet_twsk_put(tw);
134 			return TCP_TW_SUCCESS;
135 		}
136 
137 		/* New data or FIN. If new data arrive after half-duplex close,
138 		 * reset.
139 		 */
140 		if (!th->fin ||
141 		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
142 			return TCP_TW_RST;
143 
144 		/* FIN arrived, enter true time-wait state. */
145 		tw->tw_substate	  = TCP_TIME_WAIT;
146 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
147 		if (tmp_opt.saw_tstamp) {
148 			tcptw->tw_ts_recent_stamp = get_seconds();
149 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
150 		}
151 
152 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
153 		return TCP_TW_ACK;
154 	}
155 
156 	/*
157 	 *	Now real TIME-WAIT state.
158 	 *
159 	 *	RFC 1122:
160 	 *	"When a connection is [...] on TIME-WAIT state [...]
161 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
162 	 *	reopen the connection directly, if it:
163 	 *
164 	 *	(1)  assigns its initial sequence number for the new
165 	 *	connection to be larger than the largest sequence
166 	 *	number it used on the previous connection incarnation,
167 	 *	and
168 	 *
169 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
170 	 *	to be an old duplicate".
171 	 */
172 
173 	if (!paws_reject &&
174 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
175 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
176 		/* In window segment, it may be only reset or bare ack. */
177 
178 		if (th->rst) {
179 			/* This is TIME_WAIT assassination, in two flavors.
180 			 * Oh well... nobody has a sufficient solution to this
181 			 * protocol bug yet.
182 			 */
183 			if (sysctl_tcp_rfc1337 == 0) {
184 kill:
185 				inet_twsk_deschedule_put(tw);
186 				return TCP_TW_SUCCESS;
187 			}
188 		}
189 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
190 
191 		if (tmp_opt.saw_tstamp) {
192 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
193 			tcptw->tw_ts_recent_stamp = get_seconds();
194 		}
195 
196 		inet_twsk_put(tw);
197 		return TCP_TW_SUCCESS;
198 	}
199 
200 	/* Out of window segment.
201 
202 	   All the segments are ACKed immediately.
203 
204 	   The only exception is new SYN. We accept it, if it is
205 	   not old duplicate and we are not in danger to be killed
206 	   by delayed old duplicates. RFC check is that it has
207 	   newer sequence number works at rates <40Mbit/sec.
208 	   However, if paws works, it is reliable AND even more,
209 	   we even may relax silly seq space cutoff.
210 
211 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
212 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
213 	   we must return socket to time-wait state. It is not good,
214 	   but not fatal yet.
215 	 */
216 
217 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
218 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
219 	     (tmp_opt.saw_tstamp &&
220 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
221 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
222 		if (isn == 0)
223 			isn++;
224 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
225 		return TCP_TW_SYN;
226 	}
227 
228 	if (paws_reject)
229 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
230 
231 	if (!th->rst) {
232 		/* In this case we must reset the TIMEWAIT timer.
233 		 *
234 		 * If it is ACKless SYN it may be both old duplicate
235 		 * and new good SYN with random sequence number <rcv_nxt.
236 		 * Do not reschedule in the last case.
237 		 */
238 		if (paws_reject || th->ack)
239 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
240 
241 		return tcp_timewait_check_oow_rate_limit(
242 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
243 	}
244 	inet_twsk_put(tw);
245 	return TCP_TW_SUCCESS;
246 }
247 EXPORT_SYMBOL(tcp_timewait_state_process);
248 
249 /*
250  * Move a socket to time-wait or dead fin-wait-2 state.
251  */
252 void tcp_time_wait(struct sock *sk, int state, int timeo)
253 {
254 	const struct inet_connection_sock *icsk = inet_csk(sk);
255 	const struct tcp_sock *tp = tcp_sk(sk);
256 	struct inet_timewait_sock *tw;
257 	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
258 
259 	tw = inet_twsk_alloc(sk, tcp_death_row, state);
260 
261 	if (tw) {
262 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
263 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
264 		struct inet_sock *inet = inet_sk(sk);
265 
266 		tw->tw_transparent	= inet->transparent;
267 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
268 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
269 		tcptw->tw_snd_nxt	= tp->snd_nxt;
270 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
271 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
272 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
273 		tcptw->tw_ts_offset	= tp->tsoffset;
274 		tcptw->tw_last_oow_ack_time = 0;
275 
276 #if IS_ENABLED(CONFIG_IPV6)
277 		if (tw->tw_family == PF_INET6) {
278 			struct ipv6_pinfo *np = inet6_sk(sk);
279 
280 			tw->tw_v6_daddr = sk->sk_v6_daddr;
281 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
282 			tw->tw_tclass = np->tclass;
283 			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
284 			tw->tw_ipv6only = sk->sk_ipv6only;
285 		}
286 #endif
287 
288 #ifdef CONFIG_TCP_MD5SIG
289 		/*
290 		 * The timewait bucket does not have the key DB from the
291 		 * sock structure. We just make a quick copy of the
292 		 * md5 key being used (if indeed we are using one)
293 		 * so the timewait ack generating code has the key.
294 		 */
295 		do {
296 			struct tcp_md5sig_key *key;
297 			tcptw->tw_md5_key = NULL;
298 			key = tp->af_specific->md5_lookup(sk, sk);
299 			if (key) {
300 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
301 				if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
302 					BUG();
303 			}
304 		} while (0);
305 #endif
306 
307 		/* Get the TIME_WAIT timeout firing. */
308 		if (timeo < rto)
309 			timeo = rto;
310 
311 		tw->tw_timeout = TCP_TIMEWAIT_LEN;
312 		if (state == TCP_TIME_WAIT)
313 			timeo = TCP_TIMEWAIT_LEN;
314 
315 		inet_twsk_schedule(tw, timeo);
316 		/* Linkage updates. */
317 		__inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
318 		inet_twsk_put(tw);
319 	} else {
320 		/* Sorry, if we're out of memory, just CLOSE this
321 		 * socket up.  We've got bigger problems than
322 		 * non-graceful socket closings.
323 		 */
324 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
325 	}
326 
327 	tcp_update_metrics(sk);
328 	tcp_done(sk);
329 }
330 
331 void tcp_twsk_destructor(struct sock *sk)
332 {
333 #ifdef CONFIG_TCP_MD5SIG
334 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
335 
336 	if (twsk->tw_md5_key)
337 		kfree_rcu(twsk->tw_md5_key, rcu);
338 #endif
339 }
340 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
341 
342 /* Warning : This function is called without sk_listener being locked.
343  * Be sure to read socket fields once, as their value could change under us.
344  */
345 void tcp_openreq_init_rwin(struct request_sock *req,
346 			   const struct sock *sk_listener,
347 			   const struct dst_entry *dst)
348 {
349 	struct inet_request_sock *ireq = inet_rsk(req);
350 	const struct tcp_sock *tp = tcp_sk(sk_listener);
351 	int full_space = tcp_full_space(sk_listener);
352 	u32 window_clamp;
353 	__u8 rcv_wscale;
354 	int mss;
355 
356 	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
357 	window_clamp = READ_ONCE(tp->window_clamp);
358 	/* Set this up on the first call only */
359 	req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
360 
361 	/* limit the window selection if the user enforce a smaller rx buffer */
362 	if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
363 	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
364 		req->rsk_window_clamp = full_space;
365 
366 	/* tcp_full_space because it is guaranteed to be the first packet */
367 	tcp_select_initial_window(full_space,
368 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
369 		&req->rsk_rcv_wnd,
370 		&req->rsk_window_clamp,
371 		ireq->wscale_ok,
372 		&rcv_wscale,
373 		dst_metric(dst, RTAX_INITRWND));
374 	ireq->rcv_wscale = rcv_wscale;
375 }
376 EXPORT_SYMBOL(tcp_openreq_init_rwin);
377 
378 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
379 				  const struct request_sock *req)
380 {
381 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
382 }
383 
384 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
385 {
386 	struct inet_connection_sock *icsk = inet_csk(sk);
387 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
388 	bool ca_got_dst = false;
389 
390 	if (ca_key != TCP_CA_UNSPEC) {
391 		const struct tcp_congestion_ops *ca;
392 
393 		rcu_read_lock();
394 		ca = tcp_ca_find_key(ca_key);
395 		if (likely(ca && try_module_get(ca->owner))) {
396 			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
397 			icsk->icsk_ca_ops = ca;
398 			ca_got_dst = true;
399 		}
400 		rcu_read_unlock();
401 	}
402 
403 	/* If no valid choice made yet, assign current system default ca. */
404 	if (!ca_got_dst &&
405 	    (!icsk->icsk_ca_setsockopt ||
406 	     !try_module_get(icsk->icsk_ca_ops->owner)))
407 		tcp_assign_congestion_control(sk);
408 
409 	tcp_set_ca_state(sk, TCP_CA_Open);
410 }
411 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
412 
413 /* This is not only more efficient than what we used to do, it eliminates
414  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
415  *
416  * Actually, we could lots of memory writes here. tp of listening
417  * socket contains all necessary default parameters.
418  */
419 struct sock *tcp_create_openreq_child(const struct sock *sk,
420 				      struct request_sock *req,
421 				      struct sk_buff *skb)
422 {
423 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
424 
425 	if (newsk) {
426 		const struct inet_request_sock *ireq = inet_rsk(req);
427 		struct tcp_request_sock *treq = tcp_rsk(req);
428 		struct inet_connection_sock *newicsk = inet_csk(newsk);
429 		struct tcp_sock *newtp = tcp_sk(newsk);
430 
431 		/* Now setup tcp_sock */
432 		newtp->pred_flags = 0;
433 
434 		newtp->rcv_wup = newtp->copied_seq =
435 		newtp->rcv_nxt = treq->rcv_isn + 1;
436 		newtp->segs_in = 1;
437 
438 		newtp->snd_sml = newtp->snd_una =
439 		newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
440 
441 		tcp_prequeue_init(newtp);
442 		INIT_LIST_HEAD(&newtp->tsq_node);
443 
444 		tcp_init_wl(newtp, treq->rcv_isn);
445 
446 		newtp->srtt_us = 0;
447 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
448 		minmax_reset(&newtp->rtt_min, tcp_time_stamp, ~0U);
449 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
450 		newicsk->icsk_ack.lrcvtime = tcp_time_stamp;
451 
452 		newtp->packets_out = 0;
453 		newtp->retrans_out = 0;
454 		newtp->sacked_out = 0;
455 		newtp->fackets_out = 0;
456 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
457 		newtp->tlp_high_seq = 0;
458 		newtp->lsndtime = treq->snt_synack.stamp_jiffies;
459 		newsk->sk_txhash = treq->txhash;
460 		newtp->last_oow_ack_time = 0;
461 		newtp->total_retrans = req->num_retrans;
462 
463 		/* So many TCP implementations out there (incorrectly) count the
464 		 * initial SYN frame in their delayed-ACK and congestion control
465 		 * algorithms that we must have the following bandaid to talk
466 		 * efficiently to them.  -DaveM
467 		 */
468 		newtp->snd_cwnd = TCP_INIT_CWND;
469 		newtp->snd_cwnd_cnt = 0;
470 
471 		/* There's a bubble in the pipe until at least the first ACK. */
472 		newtp->app_limited = ~0U;
473 
474 		tcp_init_xmit_timers(newsk);
475 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
476 
477 		newtp->rx_opt.saw_tstamp = 0;
478 
479 		newtp->rx_opt.dsack = 0;
480 		newtp->rx_opt.num_sacks = 0;
481 
482 		newtp->urg_data = 0;
483 
484 		if (sock_flag(newsk, SOCK_KEEPOPEN))
485 			inet_csk_reset_keepalive_timer(newsk,
486 						       keepalive_time_when(newtp));
487 
488 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
489 		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
490 			if (sysctl_tcp_fack)
491 				tcp_enable_fack(newtp);
492 		}
493 		newtp->window_clamp = req->rsk_window_clamp;
494 		newtp->rcv_ssthresh = req->rsk_rcv_wnd;
495 		newtp->rcv_wnd = req->rsk_rcv_wnd;
496 		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
497 		if (newtp->rx_opt.wscale_ok) {
498 			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
499 			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
500 		} else {
501 			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
502 			newtp->window_clamp = min(newtp->window_clamp, 65535U);
503 		}
504 		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
505 				  newtp->rx_opt.snd_wscale);
506 		newtp->max_window = newtp->snd_wnd;
507 
508 		if (newtp->rx_opt.tstamp_ok) {
509 			newtp->rx_opt.ts_recent = req->ts_recent;
510 			newtp->rx_opt.ts_recent_stamp = get_seconds();
511 			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
512 		} else {
513 			newtp->rx_opt.ts_recent_stamp = 0;
514 			newtp->tcp_header_len = sizeof(struct tcphdr);
515 		}
516 		newtp->tsoffset = treq->ts_off;
517 #ifdef CONFIG_TCP_MD5SIG
518 		newtp->md5sig_info = NULL;	/*XXX*/
519 		if (newtp->af_specific->md5_lookup(sk, newsk))
520 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
521 #endif
522 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
523 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
524 		newtp->rx_opt.mss_clamp = req->mss;
525 		tcp_ecn_openreq_child(newtp, req);
526 		newtp->fastopen_req = NULL;
527 		newtp->fastopen_rsk = NULL;
528 		newtp->syn_data_acked = 0;
529 		newtp->rack.mstamp.v64 = 0;
530 		newtp->rack.advanced = 0;
531 
532 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
533 	}
534 	return newsk;
535 }
536 EXPORT_SYMBOL(tcp_create_openreq_child);
537 
538 /*
539  * Process an incoming packet for SYN_RECV sockets represented as a
540  * request_sock. Normally sk is the listener socket but for TFO it
541  * points to the child socket.
542  *
543  * XXX (TFO) - The current impl contains a special check for ack
544  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
545  *
546  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
547  */
548 
549 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
550 			   struct request_sock *req,
551 			   bool fastopen)
552 {
553 	struct tcp_options_received tmp_opt;
554 	struct sock *child;
555 	const struct tcphdr *th = tcp_hdr(skb);
556 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
557 	bool paws_reject = false;
558 	bool own_req;
559 
560 	tmp_opt.saw_tstamp = 0;
561 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
562 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
563 
564 		if (tmp_opt.saw_tstamp) {
565 			tmp_opt.ts_recent = req->ts_recent;
566 			if (tmp_opt.rcv_tsecr)
567 				tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
568 			/* We do not store true stamp, but it is not required,
569 			 * it can be estimated (approximately)
570 			 * from another data.
571 			 */
572 			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
573 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
574 		}
575 	}
576 
577 	/* Check for pure retransmitted SYN. */
578 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
579 	    flg == TCP_FLAG_SYN &&
580 	    !paws_reject) {
581 		/*
582 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
583 		 * this case on figure 6 and figure 8, but formal
584 		 * protocol description says NOTHING.
585 		 * To be more exact, it says that we should send ACK,
586 		 * because this segment (at least, if it has no data)
587 		 * is out of window.
588 		 *
589 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
590 		 *  describe SYN-RECV state. All the description
591 		 *  is wrong, we cannot believe to it and should
592 		 *  rely only on common sense and implementation
593 		 *  experience.
594 		 *
595 		 * Enforce "SYN-ACK" according to figure 8, figure 6
596 		 * of RFC793, fixed by RFC1122.
597 		 *
598 		 * Note that even if there is new data in the SYN packet
599 		 * they will be thrown away too.
600 		 *
601 		 * Reset timer after retransmitting SYNACK, similar to
602 		 * the idea of fast retransmit in recovery.
603 		 */
604 		if (!tcp_oow_rate_limited(sock_net(sk), skb,
605 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
606 					  &tcp_rsk(req)->last_oow_ack_time) &&
607 
608 		    !inet_rtx_syn_ack(sk, req)) {
609 			unsigned long expires = jiffies;
610 
611 			expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
612 				       TCP_RTO_MAX);
613 			if (!fastopen)
614 				mod_timer_pending(&req->rsk_timer, expires);
615 			else
616 				req->rsk_timer.expires = expires;
617 		}
618 		return NULL;
619 	}
620 
621 	/* Further reproduces section "SEGMENT ARRIVES"
622 	   for state SYN-RECEIVED of RFC793.
623 	   It is broken, however, it does not work only
624 	   when SYNs are crossed.
625 
626 	   You would think that SYN crossing is impossible here, since
627 	   we should have a SYN_SENT socket (from connect()) on our end,
628 	   but this is not true if the crossed SYNs were sent to both
629 	   ends by a malicious third party.  We must defend against this,
630 	   and to do that we first verify the ACK (as per RFC793, page
631 	   36) and reset if it is invalid.  Is this a true full defense?
632 	   To convince ourselves, let us consider a way in which the ACK
633 	   test can still pass in this 'malicious crossed SYNs' case.
634 	   Malicious sender sends identical SYNs (and thus identical sequence
635 	   numbers) to both A and B:
636 
637 		A: gets SYN, seq=7
638 		B: gets SYN, seq=7
639 
640 	   By our good fortune, both A and B select the same initial
641 	   send sequence number of seven :-)
642 
643 		A: sends SYN|ACK, seq=7, ack_seq=8
644 		B: sends SYN|ACK, seq=7, ack_seq=8
645 
646 	   So we are now A eating this SYN|ACK, ACK test passes.  So
647 	   does sequence test, SYN is truncated, and thus we consider
648 	   it a bare ACK.
649 
650 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
651 	   bare ACK.  Otherwise, we create an established connection.  Both
652 	   ends (listening sockets) accept the new incoming connection and try
653 	   to talk to each other. 8-)
654 
655 	   Note: This case is both harmless, and rare.  Possibility is about the
656 	   same as us discovering intelligent life on another plant tomorrow.
657 
658 	   But generally, we should (RFC lies!) to accept ACK
659 	   from SYNACK both here and in tcp_rcv_state_process().
660 	   tcp_rcv_state_process() does not, hence, we do not too.
661 
662 	   Note that the case is absolutely generic:
663 	   we cannot optimize anything here without
664 	   violating protocol. All the checks must be made
665 	   before attempt to create socket.
666 	 */
667 
668 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
669 	 *                  and the incoming segment acknowledges something not yet
670 	 *                  sent (the segment carries an unacceptable ACK) ...
671 	 *                  a reset is sent."
672 	 *
673 	 * Invalid ACK: reset will be sent by listening socket.
674 	 * Note that the ACK validity check for a Fast Open socket is done
675 	 * elsewhere and is checked directly against the child socket rather
676 	 * than req because user data may have been sent out.
677 	 */
678 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
679 	    (TCP_SKB_CB(skb)->ack_seq !=
680 	     tcp_rsk(req)->snt_isn + 1))
681 		return sk;
682 
683 	/* Also, it would be not so bad idea to check rcv_tsecr, which
684 	 * is essentially ACK extension and too early or too late values
685 	 * should cause reset in unsynchronized states.
686 	 */
687 
688 	/* RFC793: "first check sequence number". */
689 
690 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
691 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
692 		/* Out of window: send ACK and drop. */
693 		if (!(flg & TCP_FLAG_RST) &&
694 		    !tcp_oow_rate_limited(sock_net(sk), skb,
695 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
696 					  &tcp_rsk(req)->last_oow_ack_time))
697 			req->rsk_ops->send_ack(sk, skb, req);
698 		if (paws_reject)
699 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
700 		return NULL;
701 	}
702 
703 	/* In sequence, PAWS is OK. */
704 
705 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
706 		req->ts_recent = tmp_opt.rcv_tsval;
707 
708 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
709 		/* Truncate SYN, it is out of window starting
710 		   at tcp_rsk(req)->rcv_isn + 1. */
711 		flg &= ~TCP_FLAG_SYN;
712 	}
713 
714 	/* RFC793: "second check the RST bit" and
715 	 *	   "fourth, check the SYN bit"
716 	 */
717 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
718 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
719 		goto embryonic_reset;
720 	}
721 
722 	/* ACK sequence verified above, just make sure ACK is
723 	 * set.  If ACK not set, just silently drop the packet.
724 	 *
725 	 * XXX (TFO) - if we ever allow "data after SYN", the
726 	 * following check needs to be removed.
727 	 */
728 	if (!(flg & TCP_FLAG_ACK))
729 		return NULL;
730 
731 	/* For Fast Open no more processing is needed (sk is the
732 	 * child socket).
733 	 */
734 	if (fastopen)
735 		return sk;
736 
737 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
738 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
739 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
740 		inet_rsk(req)->acked = 1;
741 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
742 		return NULL;
743 	}
744 
745 	/* OK, ACK is valid, create big socket and
746 	 * feed this segment to it. It will repeat all
747 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
748 	 * ESTABLISHED STATE. If it will be dropped after
749 	 * socket is created, wait for troubles.
750 	 */
751 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
752 							 req, &own_req);
753 	if (!child)
754 		goto listen_overflow;
755 
756 	sock_rps_save_rxhash(child, skb);
757 	tcp_synack_rtt_meas(child, req);
758 	return inet_csk_complete_hashdance(sk, child, req, own_req);
759 
760 listen_overflow:
761 	if (!sysctl_tcp_abort_on_overflow) {
762 		inet_rsk(req)->acked = 1;
763 		return NULL;
764 	}
765 
766 embryonic_reset:
767 	if (!(flg & TCP_FLAG_RST)) {
768 		/* Received a bad SYN pkt - for TFO We try not to reset
769 		 * the local connection unless it's really necessary to
770 		 * avoid becoming vulnerable to outside attack aiming at
771 		 * resetting legit local connections.
772 		 */
773 		req->rsk_ops->send_reset(sk, skb);
774 	} else if (fastopen) { /* received a valid RST pkt */
775 		reqsk_fastopen_remove(sk, req, true);
776 		tcp_reset(sk);
777 	}
778 	if (!fastopen) {
779 		inet_csk_reqsk_queue_drop(sk, req);
780 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
781 	}
782 	return NULL;
783 }
784 EXPORT_SYMBOL(tcp_check_req);
785 
786 /*
787  * Queue segment on the new socket if the new socket is active,
788  * otherwise we just shortcircuit this and continue with
789  * the new socket.
790  *
791  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
792  * when entering. But other states are possible due to a race condition
793  * where after __inet_lookup_established() fails but before the listener
794  * locked is obtained, other packets cause the same connection to
795  * be created.
796  */
797 
798 int tcp_child_process(struct sock *parent, struct sock *child,
799 		      struct sk_buff *skb)
800 {
801 	int ret = 0;
802 	int state = child->sk_state;
803 
804 	/* record NAPI ID of child */
805 	sk_mark_napi_id(child, skb);
806 
807 	tcp_segs_in(tcp_sk(child), skb);
808 	if (!sock_owned_by_user(child)) {
809 		ret = tcp_rcv_state_process(child, skb);
810 		/* Wakeup parent, send SIGIO */
811 		if (state == TCP_SYN_RECV && child->sk_state != state)
812 			parent->sk_data_ready(parent);
813 	} else {
814 		/* Alas, it is possible again, because we do lookup
815 		 * in main socket hash table and lock on listening
816 		 * socket does not protect us more.
817 		 */
818 		__sk_add_backlog(child, skb);
819 	}
820 
821 	bh_unlock_sock(child);
822 	sock_put(child);
823 	return ret;
824 }
825 EXPORT_SYMBOL(tcp_child_process);
826