xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision a2cce7a9)
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 	.hashinfo	= &tcp_hashinfo,
38 };
39 EXPORT_SYMBOL_GPL(tcp_death_row);
40 
41 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
42 {
43 	if (seq == s_win)
44 		return true;
45 	if (after(end_seq, s_win) && before(seq, e_win))
46 		return true;
47 	return seq == e_win && seq == end_seq;
48 }
49 
50 static enum tcp_tw_status
51 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
52 				  const struct sk_buff *skb, int mib_idx)
53 {
54 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
55 
56 	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
57 				  &tcptw->tw_last_oow_ack_time)) {
58 		/* Send ACK. Note, we do not put the bucket,
59 		 * it will be released by caller.
60 		 */
61 		return TCP_TW_ACK;
62 	}
63 
64 	/* We are rate-limiting, so just release the tw sock and drop skb. */
65 	inet_twsk_put(tw);
66 	return TCP_TW_SUCCESS;
67 }
68 
69 /*
70  * * Main purpose of TIME-WAIT state is to close connection gracefully,
71  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
72  *   (and, probably, tail of data) and one or more our ACKs are lost.
73  * * What is TIME-WAIT timeout? It is associated with maximal packet
74  *   lifetime in the internet, which results in wrong conclusion, that
75  *   it is set to catch "old duplicate segments" wandering out of their path.
76  *   It is not quite correct. This timeout is calculated so that it exceeds
77  *   maximal retransmission timeout enough to allow to lose one (or more)
78  *   segments sent by peer and our ACKs. This time may be calculated from RTO.
79  * * When TIME-WAIT socket receives RST, it means that another end
80  *   finally closed and we are allowed to kill TIME-WAIT too.
81  * * Second purpose of TIME-WAIT is catching old duplicate segments.
82  *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
83  *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
84  * * If we invented some more clever way to catch duplicates
85  *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
86  *
87  * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
88  * When you compare it to RFCs, please, read section SEGMENT ARRIVES
89  * from the very beginning.
90  *
91  * NOTE. With recycling (and later with fin-wait-2) TW bucket
92  * is _not_ stateless. It means, that strictly speaking we must
93  * spinlock it. I do not want! Well, probability of misbehaviour
94  * is ridiculously low and, seems, we could use some mb() tricks
95  * to avoid misread sequence numbers, states etc.  --ANK
96  *
97  * We don't need to initialize tmp_out.sack_ok as we don't use the results
98  */
99 enum tcp_tw_status
100 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
101 			   const struct tcphdr *th)
102 {
103 	struct tcp_options_received tmp_opt;
104 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
105 	bool paws_reject = false;
106 
107 	tmp_opt.saw_tstamp = 0;
108 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
109 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
110 
111 		if (tmp_opt.saw_tstamp) {
112 			tmp_opt.rcv_tsecr	-= tcptw->tw_ts_offset;
113 			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
114 			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
115 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
116 		}
117 	}
118 
119 	if (tw->tw_substate == TCP_FIN_WAIT2) {
120 		/* Just repeat all the checks of tcp_rcv_state_process() */
121 
122 		/* Out of window, send ACK */
123 		if (paws_reject ||
124 		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
125 				   tcptw->tw_rcv_nxt,
126 				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
127 			return tcp_timewait_check_oow_rate_limit(
128 				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
129 
130 		if (th->rst)
131 			goto kill;
132 
133 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
134 			goto kill_with_rst;
135 
136 		/* Dup ACK? */
137 		if (!th->ack ||
138 		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
139 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
140 			inet_twsk_put(tw);
141 			return TCP_TW_SUCCESS;
142 		}
143 
144 		/* New data or FIN. If new data arrive after half-duplex close,
145 		 * reset.
146 		 */
147 		if (!th->fin ||
148 		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
149 kill_with_rst:
150 			inet_twsk_deschedule_put(tw);
151 			return TCP_TW_RST;
152 		}
153 
154 		/* FIN arrived, enter true time-wait state. */
155 		tw->tw_substate	  = TCP_TIME_WAIT;
156 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
157 		if (tmp_opt.saw_tstamp) {
158 			tcptw->tw_ts_recent_stamp = get_seconds();
159 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
160 		}
161 
162 		if (tcp_death_row.sysctl_tw_recycle &&
163 		    tcptw->tw_ts_recent_stamp &&
164 		    tcp_tw_remember_stamp(tw))
165 			inet_twsk_reschedule(tw, tw->tw_timeout);
166 		else
167 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
168 		return TCP_TW_ACK;
169 	}
170 
171 	/*
172 	 *	Now real TIME-WAIT state.
173 	 *
174 	 *	RFC 1122:
175 	 *	"When a connection is [...] on TIME-WAIT state [...]
176 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
177 	 *	reopen the connection directly, if it:
178 	 *
179 	 *	(1)  assigns its initial sequence number for the new
180 	 *	connection to be larger than the largest sequence
181 	 *	number it used on the previous connection incarnation,
182 	 *	and
183 	 *
184 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
185 	 *	to be an old duplicate".
186 	 */
187 
188 	if (!paws_reject &&
189 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
190 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
191 		/* In window segment, it may be only reset or bare ack. */
192 
193 		if (th->rst) {
194 			/* This is TIME_WAIT assassination, in two flavors.
195 			 * Oh well... nobody has a sufficient solution to this
196 			 * protocol bug yet.
197 			 */
198 			if (sysctl_tcp_rfc1337 == 0) {
199 kill:
200 				inet_twsk_deschedule_put(tw);
201 				return TCP_TW_SUCCESS;
202 			}
203 		}
204 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
205 
206 		if (tmp_opt.saw_tstamp) {
207 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
208 			tcptw->tw_ts_recent_stamp = get_seconds();
209 		}
210 
211 		inet_twsk_put(tw);
212 		return TCP_TW_SUCCESS;
213 	}
214 
215 	/* Out of window segment.
216 
217 	   All the segments are ACKed immediately.
218 
219 	   The only exception is new SYN. We accept it, if it is
220 	   not old duplicate and we are not in danger to be killed
221 	   by delayed old duplicates. RFC check is that it has
222 	   newer sequence number works at rates <40Mbit/sec.
223 	   However, if paws works, it is reliable AND even more,
224 	   we even may relax silly seq space cutoff.
225 
226 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
227 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
228 	   we must return socket to time-wait state. It is not good,
229 	   but not fatal yet.
230 	 */
231 
232 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
233 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
234 	     (tmp_opt.saw_tstamp &&
235 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
236 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
237 		if (isn == 0)
238 			isn++;
239 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
240 		return TCP_TW_SYN;
241 	}
242 
243 	if (paws_reject)
244 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
245 
246 	if (!th->rst) {
247 		/* In this case we must reset the TIMEWAIT timer.
248 		 *
249 		 * If it is ACKless SYN it may be both old duplicate
250 		 * and new good SYN with random sequence number <rcv_nxt.
251 		 * Do not reschedule in the last case.
252 		 */
253 		if (paws_reject || th->ack)
254 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
255 
256 		return tcp_timewait_check_oow_rate_limit(
257 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
258 	}
259 	inet_twsk_put(tw);
260 	return TCP_TW_SUCCESS;
261 }
262 EXPORT_SYMBOL(tcp_timewait_state_process);
263 
264 /*
265  * Move a socket to time-wait or dead fin-wait-2 state.
266  */
267 void tcp_time_wait(struct sock *sk, int state, int timeo)
268 {
269 	const struct inet_connection_sock *icsk = inet_csk(sk);
270 	const struct tcp_sock *tp = tcp_sk(sk);
271 	struct inet_timewait_sock *tw;
272 	bool recycle_ok = false;
273 
274 	if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
275 		recycle_ok = tcp_remember_stamp(sk);
276 
277 	tw = inet_twsk_alloc(sk, &tcp_death_row, state);
278 
279 	if (tw) {
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 		tcptw->tw_last_oow_ack_time = 0;
293 
294 #if IS_ENABLED(CONFIG_IPV6)
295 		if (tw->tw_family == PF_INET6) {
296 			struct ipv6_pinfo *np = inet6_sk(sk);
297 
298 			tw->tw_v6_daddr = sk->sk_v6_daddr;
299 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
300 			tw->tw_tclass = np->tclass;
301 			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
302 			tw->tw_ipv6only = sk->sk_ipv6only;
303 		}
304 #endif
305 
306 #ifdef CONFIG_TCP_MD5SIG
307 		/*
308 		 * The timewait bucket does not have the key DB from the
309 		 * sock structure. We just make a quick copy of the
310 		 * md5 key being used (if indeed we are using one)
311 		 * so the timewait ack generating code has the key.
312 		 */
313 		do {
314 			struct tcp_md5sig_key *key;
315 			tcptw->tw_md5_key = NULL;
316 			key = tp->af_specific->md5_lookup(sk, sk);
317 			if (key) {
318 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
319 				if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
320 					BUG();
321 			}
322 		} while (0);
323 #endif
324 
325 		/* Get the TIME_WAIT timeout firing. */
326 		if (timeo < rto)
327 			timeo = rto;
328 
329 		if (recycle_ok) {
330 			tw->tw_timeout = rto;
331 		} else {
332 			tw->tw_timeout = TCP_TIMEWAIT_LEN;
333 			if (state == TCP_TIME_WAIT)
334 				timeo = TCP_TIMEWAIT_LEN;
335 		}
336 
337 		inet_twsk_schedule(tw, timeo);
338 		/* Linkage updates. */
339 		__inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
340 		inet_twsk_put(tw);
341 	} else {
342 		/* Sorry, if we're out of memory, just CLOSE this
343 		 * socket up.  We've got bigger problems than
344 		 * non-graceful socket closings.
345 		 */
346 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
347 	}
348 
349 	tcp_update_metrics(sk);
350 	tcp_done(sk);
351 }
352 
353 void tcp_twsk_destructor(struct sock *sk)
354 {
355 #ifdef CONFIG_TCP_MD5SIG
356 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
357 
358 	if (twsk->tw_md5_key)
359 		kfree_rcu(twsk->tw_md5_key, rcu);
360 #endif
361 }
362 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
363 
364 void tcp_openreq_init_rwin(struct request_sock *req,
365 			   struct sock *sk, struct dst_entry *dst)
366 {
367 	struct inet_request_sock *ireq = inet_rsk(req);
368 	struct tcp_sock *tp = tcp_sk(sk);
369 	__u8 rcv_wscale;
370 	int mss = dst_metric_advmss(dst);
371 
372 	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
373 		mss = tp->rx_opt.user_mss;
374 
375 	/* Set this up on the first call only */
376 	req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
377 
378 	/* limit the window selection if the user enforce a smaller rx buffer */
379 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
380 	    (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0))
381 		req->window_clamp = tcp_full_space(sk);
382 
383 	/* tcp_full_space because it is guaranteed to be the first packet */
384 	tcp_select_initial_window(tcp_full_space(sk),
385 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
386 		&req->rcv_wnd,
387 		&req->window_clamp,
388 		ireq->wscale_ok,
389 		&rcv_wscale,
390 		dst_metric(dst, RTAX_INITRWND));
391 	ireq->rcv_wscale = rcv_wscale;
392 }
393 EXPORT_SYMBOL(tcp_openreq_init_rwin);
394 
395 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
396 				  const struct request_sock *req)
397 {
398 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
399 }
400 
401 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
402 {
403 	struct inet_connection_sock *icsk = inet_csk(sk);
404 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
405 	bool ca_got_dst = false;
406 
407 	if (ca_key != TCP_CA_UNSPEC) {
408 		const struct tcp_congestion_ops *ca;
409 
410 		rcu_read_lock();
411 		ca = tcp_ca_find_key(ca_key);
412 		if (likely(ca && try_module_get(ca->owner))) {
413 			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
414 			icsk->icsk_ca_ops = ca;
415 			ca_got_dst = true;
416 		}
417 		rcu_read_unlock();
418 	}
419 
420 	/* If no valid choice made yet, assign current system default ca. */
421 	if (!ca_got_dst &&
422 	    (!icsk->icsk_ca_setsockopt ||
423 	     !try_module_get(icsk->icsk_ca_ops->owner)))
424 		tcp_assign_congestion_control(sk);
425 
426 	tcp_set_ca_state(sk, TCP_CA_Open);
427 }
428 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
429 
430 /* This is not only more efficient than what we used to do, it eliminates
431  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
432  *
433  * Actually, we could lots of memory writes here. tp of listening
434  * socket contains all necessary default parameters.
435  */
436 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
437 {
438 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
439 
440 	if (newsk) {
441 		const struct inet_request_sock *ireq = inet_rsk(req);
442 		struct tcp_request_sock *treq = tcp_rsk(req);
443 		struct inet_connection_sock *newicsk = inet_csk(newsk);
444 		struct tcp_sock *newtp = tcp_sk(newsk);
445 
446 		/* Now setup tcp_sock */
447 		newtp->pred_flags = 0;
448 
449 		newtp->rcv_wup = newtp->copied_seq =
450 		newtp->rcv_nxt = treq->rcv_isn + 1;
451 		newtp->segs_in = 0;
452 
453 		newtp->snd_sml = newtp->snd_una =
454 		newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
455 
456 		tcp_prequeue_init(newtp);
457 		INIT_LIST_HEAD(&newtp->tsq_node);
458 
459 		tcp_init_wl(newtp, treq->rcv_isn);
460 
461 		newtp->srtt_us = 0;
462 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
463 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
464 
465 		newtp->packets_out = 0;
466 		newtp->retrans_out = 0;
467 		newtp->sacked_out = 0;
468 		newtp->fackets_out = 0;
469 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
470 		tcp_enable_early_retrans(newtp);
471 		newtp->tlp_high_seq = 0;
472 		newtp->lsndtime = treq->snt_synack;
473 		newtp->last_oow_ack_time = 0;
474 		newtp->total_retrans = req->num_retrans;
475 
476 		/* So many TCP implementations out there (incorrectly) count the
477 		 * initial SYN frame in their delayed-ACK and congestion control
478 		 * algorithms that we must have the following bandaid to talk
479 		 * efficiently to them.  -DaveM
480 		 */
481 		newtp->snd_cwnd = TCP_INIT_CWND;
482 		newtp->snd_cwnd_cnt = 0;
483 
484 		tcp_init_xmit_timers(newsk);
485 		__skb_queue_head_init(&newtp->out_of_order_queue);
486 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
487 
488 		newtp->rx_opt.saw_tstamp = 0;
489 
490 		newtp->rx_opt.dsack = 0;
491 		newtp->rx_opt.num_sacks = 0;
492 
493 		newtp->urg_data = 0;
494 
495 		if (sock_flag(newsk, SOCK_KEEPOPEN))
496 			inet_csk_reset_keepalive_timer(newsk,
497 						       keepalive_time_when(newtp));
498 
499 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
500 		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
501 			if (sysctl_tcp_fack)
502 				tcp_enable_fack(newtp);
503 		}
504 		newtp->window_clamp = req->window_clamp;
505 		newtp->rcv_ssthresh = req->rcv_wnd;
506 		newtp->rcv_wnd = req->rcv_wnd;
507 		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
508 		if (newtp->rx_opt.wscale_ok) {
509 			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
510 			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
511 		} else {
512 			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
513 			newtp->window_clamp = min(newtp->window_clamp, 65535U);
514 		}
515 		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
516 				  newtp->rx_opt.snd_wscale);
517 		newtp->max_window = newtp->snd_wnd;
518 
519 		if (newtp->rx_opt.tstamp_ok) {
520 			newtp->rx_opt.ts_recent = req->ts_recent;
521 			newtp->rx_opt.ts_recent_stamp = get_seconds();
522 			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
523 		} else {
524 			newtp->rx_opt.ts_recent_stamp = 0;
525 			newtp->tcp_header_len = sizeof(struct tcphdr);
526 		}
527 		newtp->tsoffset = 0;
528 #ifdef CONFIG_TCP_MD5SIG
529 		newtp->md5sig_info = NULL;	/*XXX*/
530 		if (newtp->af_specific->md5_lookup(sk, newsk))
531 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
532 #endif
533 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
534 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
535 		newtp->rx_opt.mss_clamp = req->mss;
536 		tcp_ecn_openreq_child(newtp, req);
537 		newtp->fastopen_rsk = NULL;
538 		newtp->syn_data_acked = 0;
539 
540 		newtp->saved_syn = req->saved_syn;
541 		req->saved_syn = NULL;
542 
543 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
544 	}
545 	return newsk;
546 }
547 EXPORT_SYMBOL(tcp_create_openreq_child);
548 
549 /*
550  * Process an incoming packet for SYN_RECV sockets represented as a
551  * request_sock. Normally sk is the listener socket but for TFO it
552  * points to the child socket.
553  *
554  * XXX (TFO) - The current impl contains a special check for ack
555  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
556  *
557  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
558  */
559 
560 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
561 			   struct request_sock *req,
562 			   bool fastopen)
563 {
564 	struct tcp_options_received tmp_opt;
565 	struct sock *child;
566 	const struct tcphdr *th = tcp_hdr(skb);
567 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
568 	bool paws_reject = false;
569 
570 	BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
571 
572 	tmp_opt.saw_tstamp = 0;
573 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
574 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
575 
576 		if (tmp_opt.saw_tstamp) {
577 			tmp_opt.ts_recent = req->ts_recent;
578 			/* We do not store true stamp, but it is not required,
579 			 * it can be estimated (approximately)
580 			 * from another data.
581 			 */
582 			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
583 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
584 		}
585 	}
586 
587 	/* Check for pure retransmitted SYN. */
588 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
589 	    flg == TCP_FLAG_SYN &&
590 	    !paws_reject) {
591 		/*
592 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
593 		 * this case on figure 6 and figure 8, but formal
594 		 * protocol description says NOTHING.
595 		 * To be more exact, it says that we should send ACK,
596 		 * because this segment (at least, if it has no data)
597 		 * is out of window.
598 		 *
599 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
600 		 *  describe SYN-RECV state. All the description
601 		 *  is wrong, we cannot believe to it and should
602 		 *  rely only on common sense and implementation
603 		 *  experience.
604 		 *
605 		 * Enforce "SYN-ACK" according to figure 8, figure 6
606 		 * of RFC793, fixed by RFC1122.
607 		 *
608 		 * Note that even if there is new data in the SYN packet
609 		 * they will be thrown away too.
610 		 *
611 		 * Reset timer after retransmitting SYNACK, similar to
612 		 * the idea of fast retransmit in recovery.
613 		 */
614 		if (!tcp_oow_rate_limited(sock_net(sk), skb,
615 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
616 					  &tcp_rsk(req)->last_oow_ack_time) &&
617 
618 		    !inet_rtx_syn_ack(sk, req)) {
619 			unsigned long expires = jiffies;
620 
621 			expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
622 				       TCP_RTO_MAX);
623 			if (!fastopen)
624 				mod_timer_pending(&req->rsk_timer, expires);
625 			else
626 				req->rsk_timer.expires = expires;
627 		}
628 		return NULL;
629 	}
630 
631 	/* Further reproduces section "SEGMENT ARRIVES"
632 	   for state SYN-RECEIVED of RFC793.
633 	   It is broken, however, it does not work only
634 	   when SYNs are crossed.
635 
636 	   You would think that SYN crossing is impossible here, since
637 	   we should have a SYN_SENT socket (from connect()) on our end,
638 	   but this is not true if the crossed SYNs were sent to both
639 	   ends by a malicious third party.  We must defend against this,
640 	   and to do that we first verify the ACK (as per RFC793, page
641 	   36) and reset if it is invalid.  Is this a true full defense?
642 	   To convince ourselves, let us consider a way in which the ACK
643 	   test can still pass in this 'malicious crossed SYNs' case.
644 	   Malicious sender sends identical SYNs (and thus identical sequence
645 	   numbers) to both A and B:
646 
647 		A: gets SYN, seq=7
648 		B: gets SYN, seq=7
649 
650 	   By our good fortune, both A and B select the same initial
651 	   send sequence number of seven :-)
652 
653 		A: sends SYN|ACK, seq=7, ack_seq=8
654 		B: sends SYN|ACK, seq=7, ack_seq=8
655 
656 	   So we are now A eating this SYN|ACK, ACK test passes.  So
657 	   does sequence test, SYN is truncated, and thus we consider
658 	   it a bare ACK.
659 
660 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
661 	   bare ACK.  Otherwise, we create an established connection.  Both
662 	   ends (listening sockets) accept the new incoming connection and try
663 	   to talk to each other. 8-)
664 
665 	   Note: This case is both harmless, and rare.  Possibility is about the
666 	   same as us discovering intelligent life on another plant tomorrow.
667 
668 	   But generally, we should (RFC lies!) to accept ACK
669 	   from SYNACK both here and in tcp_rcv_state_process().
670 	   tcp_rcv_state_process() does not, hence, we do not too.
671 
672 	   Note that the case is absolutely generic:
673 	   we cannot optimize anything here without
674 	   violating protocol. All the checks must be made
675 	   before attempt to create socket.
676 	 */
677 
678 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
679 	 *                  and the incoming segment acknowledges something not yet
680 	 *                  sent (the segment carries an unacceptable ACK) ...
681 	 *                  a reset is sent."
682 	 *
683 	 * Invalid ACK: reset will be sent by listening socket.
684 	 * Note that the ACK validity check for a Fast Open socket is done
685 	 * elsewhere and is checked directly against the child socket rather
686 	 * than req because user data may have been sent out.
687 	 */
688 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
689 	    (TCP_SKB_CB(skb)->ack_seq !=
690 	     tcp_rsk(req)->snt_isn + 1))
691 		return sk;
692 
693 	/* Also, it would be not so bad idea to check rcv_tsecr, which
694 	 * is essentially ACK extension and too early or too late values
695 	 * should cause reset in unsynchronized states.
696 	 */
697 
698 	/* RFC793: "first check sequence number". */
699 
700 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
701 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
702 		/* Out of window: send ACK and drop. */
703 		if (!(flg & TCP_FLAG_RST))
704 			req->rsk_ops->send_ack(sk, skb, req);
705 		if (paws_reject)
706 			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
707 		return NULL;
708 	}
709 
710 	/* In sequence, PAWS is OK. */
711 
712 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
713 		req->ts_recent = tmp_opt.rcv_tsval;
714 
715 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
716 		/* Truncate SYN, it is out of window starting
717 		   at tcp_rsk(req)->rcv_isn + 1. */
718 		flg &= ~TCP_FLAG_SYN;
719 	}
720 
721 	/* RFC793: "second check the RST bit" and
722 	 *	   "fourth, check the SYN bit"
723 	 */
724 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
725 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
726 		goto embryonic_reset;
727 	}
728 
729 	/* ACK sequence verified above, just make sure ACK is
730 	 * set.  If ACK not set, just silently drop the packet.
731 	 *
732 	 * XXX (TFO) - if we ever allow "data after SYN", the
733 	 * following check needs to be removed.
734 	 */
735 	if (!(flg & TCP_FLAG_ACK))
736 		return NULL;
737 
738 	/* For Fast Open no more processing is needed (sk is the
739 	 * child socket).
740 	 */
741 	if (fastopen)
742 		return sk;
743 
744 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
745 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
746 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
747 		inet_rsk(req)->acked = 1;
748 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
749 		return NULL;
750 	}
751 
752 	/* OK, ACK is valid, create big socket and
753 	 * feed this segment to it. It will repeat all
754 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
755 	 * ESTABLISHED STATE. If it will be dropped after
756 	 * socket is created, wait for troubles.
757 	 */
758 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
759 	if (!child)
760 		goto listen_overflow;
761 
762 	inet_csk_reqsk_queue_drop(sk, req);
763 	inet_csk_reqsk_queue_add(sk, req, child);
764 	/* Warning: caller must not call reqsk_put(req);
765 	 * child stole last reference on it.
766 	 */
767 	return child;
768 
769 listen_overflow:
770 	if (!sysctl_tcp_abort_on_overflow) {
771 		inet_rsk(req)->acked = 1;
772 		return NULL;
773 	}
774 
775 embryonic_reset:
776 	if (!(flg & TCP_FLAG_RST)) {
777 		/* Received a bad SYN pkt - for TFO We try not to reset
778 		 * the local connection unless it's really necessary to
779 		 * avoid becoming vulnerable to outside attack aiming at
780 		 * resetting legit local connections.
781 		 */
782 		req->rsk_ops->send_reset(sk, skb);
783 	} else if (fastopen) { /* received a valid RST pkt */
784 		reqsk_fastopen_remove(sk, req, true);
785 		tcp_reset(sk);
786 	}
787 	if (!fastopen) {
788 		inet_csk_reqsk_queue_drop(sk, req);
789 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
790 	}
791 	return NULL;
792 }
793 EXPORT_SYMBOL(tcp_check_req);
794 
795 /*
796  * Queue segment on the new socket if the new socket is active,
797  * otherwise we just shortcircuit this and continue with
798  * the new socket.
799  *
800  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
801  * when entering. But other states are possible due to a race condition
802  * where after __inet_lookup_established() fails but before the listener
803  * locked is obtained, other packets cause the same connection to
804  * be created.
805  */
806 
807 int tcp_child_process(struct sock *parent, struct sock *child,
808 		      struct sk_buff *skb)
809 {
810 	int ret = 0;
811 	int state = child->sk_state;
812 
813 	if (!sock_owned_by_user(child)) {
814 		ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
815 					    skb->len);
816 		/* Wakeup parent, send SIGIO */
817 		if (state == TCP_SYN_RECV && child->sk_state != state)
818 			parent->sk_data_ready(parent);
819 	} else {
820 		/* Alas, it is possible again, because we do lookup
821 		 * in main socket hash table and lock on listening
822 		 * socket does not protect us more.
823 		 */
824 		__sk_add_backlog(child, skb);
825 	}
826 
827 	bh_unlock_sock(child);
828 	sock_put(child);
829 	return ret;
830 }
831 EXPORT_SYMBOL(tcp_child_process);
832