xref: /linux/net/ipv4/tcp_minisocks.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Implementation of the Transmission Control Protocol(TCP).
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
11  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
12  *		Florian La Roche, <flla@stud.uni-sb.de>
13  *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14  *		Linus Torvalds, <torvalds@cs.helsinki.fi>
15  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
16  *		Matthew Dillon, <dillon@apollo.west.oic.com>
17  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18  *		Jorge Cwik, <jorge@laser.satlink.net>
19  */
20 
21 #include <linux/mm.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <linux/sysctl.h>
25 #include <linux/workqueue.h>
26 #include <net/tcp.h>
27 #include <net/inet_common.h>
28 #include <net/xfrm.h>
29 
30 int sysctl_tcp_syncookies __read_mostly = 1;
31 EXPORT_SYMBOL(sysctl_tcp_syncookies);
32 
33 int sysctl_tcp_abort_on_overflow __read_mostly;
34 
35 struct inet_timewait_death_row tcp_death_row = {
36 	.sysctl_max_tw_buckets = NR_FILE * 2,
37 	.period		= TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
38 	.death_lock	= __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock),
39 	.hashinfo	= &tcp_hashinfo,
40 	.tw_timer	= TIMER_INITIALIZER(inet_twdr_hangman, 0,
41 					    (unsigned long)&tcp_death_row),
42 	.twkill_work	= __WORK_INITIALIZER(tcp_death_row.twkill_work,
43 					     inet_twdr_twkill_work),
44 /* Short-time timewait calendar */
45 
46 	.twcal_hand	= -1,
47 	.twcal_timer	= TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
48 					    (unsigned long)&tcp_death_row),
49 };
50 EXPORT_SYMBOL_GPL(tcp_death_row);
51 
52 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
53 {
54 	if (seq == s_win)
55 		return true;
56 	if (after(end_seq, s_win) && before(seq, e_win))
57 		return true;
58 	return seq == e_win && seq == end_seq;
59 }
60 
61 /*
62  * * Main purpose of TIME-WAIT state is to close connection gracefully,
63  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
64  *   (and, probably, tail of data) and one or more our ACKs are lost.
65  * * What is TIME-WAIT timeout? It is associated with maximal packet
66  *   lifetime in the internet, which results in wrong conclusion, that
67  *   it is set to catch "old duplicate segments" wandering out of their path.
68  *   It is not quite correct. This timeout is calculated so that it exceeds
69  *   maximal retransmission timeout enough to allow to lose one (or more)
70  *   segments sent by peer and our ACKs. This time may be calculated from RTO.
71  * * When TIME-WAIT socket receives RST, it means that another end
72  *   finally closed and we are allowed to kill TIME-WAIT too.
73  * * Second purpose of TIME-WAIT is catching old duplicate segments.
74  *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
75  *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
76  * * If we invented some more clever way to catch duplicates
77  *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
78  *
79  * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
80  * When you compare it to RFCs, please, read section SEGMENT ARRIVES
81  * from the very beginning.
82  *
83  * NOTE. With recycling (and later with fin-wait-2) TW bucket
84  * is _not_ stateless. It means, that strictly speaking we must
85  * spinlock it. I do not want! Well, probability of misbehaviour
86  * is ridiculously low and, seems, we could use some mb() tricks
87  * to avoid misread sequence numbers, states etc.  --ANK
88  *
89  * We don't need to initialize tmp_out.sack_ok as we don't use the results
90  */
91 enum tcp_tw_status
92 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
93 			   const struct tcphdr *th)
94 {
95 	struct tcp_options_received tmp_opt;
96 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
97 	bool paws_reject = false;
98 
99 	tmp_opt.saw_tstamp = 0;
100 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
101 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
102 
103 		if (tmp_opt.saw_tstamp) {
104 			tmp_opt.rcv_tsecr	-= tcptw->tw_ts_offset;
105 			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
106 			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
107 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
108 		}
109 	}
110 
111 	if (tw->tw_substate == TCP_FIN_WAIT2) {
112 		/* Just repeat all the checks of tcp_rcv_state_process() */
113 
114 		/* Out of window, send ACK */
115 		if (paws_reject ||
116 		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
117 				   tcptw->tw_rcv_nxt,
118 				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
119 			return TCP_TW_ACK;
120 
121 		if (th->rst)
122 			goto kill;
123 
124 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
125 			goto kill_with_rst;
126 
127 		/* Dup ACK? */
128 		if (!th->ack ||
129 		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
130 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
131 			inet_twsk_put(tw);
132 			return TCP_TW_SUCCESS;
133 		}
134 
135 		/* New data or FIN. If new data arrive after half-duplex close,
136 		 * reset.
137 		 */
138 		if (!th->fin ||
139 		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
140 kill_with_rst:
141 			inet_twsk_deschedule(tw, &tcp_death_row);
142 			inet_twsk_put(tw);
143 			return TCP_TW_RST;
144 		}
145 
146 		/* FIN arrived, enter true time-wait state. */
147 		tw->tw_substate	  = TCP_TIME_WAIT;
148 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
149 		if (tmp_opt.saw_tstamp) {
150 			tcptw->tw_ts_recent_stamp = get_seconds();
151 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
152 		}
153 
154 		if (tcp_death_row.sysctl_tw_recycle &&
155 		    tcptw->tw_ts_recent_stamp &&
156 		    tcp_tw_remember_stamp(tw))
157 			inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
158 					   TCP_TIMEWAIT_LEN);
159 		else
160 			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
161 					   TCP_TIMEWAIT_LEN);
162 		return TCP_TW_ACK;
163 	}
164 
165 	/*
166 	 *	Now real TIME-WAIT state.
167 	 *
168 	 *	RFC 1122:
169 	 *	"When a connection is [...] on TIME-WAIT state [...]
170 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
171 	 *	reopen the connection directly, if it:
172 	 *
173 	 *	(1)  assigns its initial sequence number for the new
174 	 *	connection to be larger than the largest sequence
175 	 *	number it used on the previous connection incarnation,
176 	 *	and
177 	 *
178 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
179 	 *	to be an old duplicate".
180 	 */
181 
182 	if (!paws_reject &&
183 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
184 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
185 		/* In window segment, it may be only reset or bare ack. */
186 
187 		if (th->rst) {
188 			/* This is TIME_WAIT assassination, in two flavors.
189 			 * Oh well... nobody has a sufficient solution to this
190 			 * protocol bug yet.
191 			 */
192 			if (sysctl_tcp_rfc1337 == 0) {
193 kill:
194 				inet_twsk_deschedule(tw, &tcp_death_row);
195 				inet_twsk_put(tw);
196 				return TCP_TW_SUCCESS;
197 			}
198 		}
199 		inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
200 				   TCP_TIMEWAIT_LEN);
201 
202 		if (tmp_opt.saw_tstamp) {
203 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
204 			tcptw->tw_ts_recent_stamp = get_seconds();
205 		}
206 
207 		inet_twsk_put(tw);
208 		return TCP_TW_SUCCESS;
209 	}
210 
211 	/* Out of window segment.
212 
213 	   All the segments are ACKed immediately.
214 
215 	   The only exception is new SYN. We accept it, if it is
216 	   not old duplicate and we are not in danger to be killed
217 	   by delayed old duplicates. RFC check is that it has
218 	   newer sequence number works at rates <40Mbit/sec.
219 	   However, if paws works, it is reliable AND even more,
220 	   we even may relax silly seq space cutoff.
221 
222 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
223 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
224 	   we must return socket to time-wait state. It is not good,
225 	   but not fatal yet.
226 	 */
227 
228 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
229 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
230 	     (tmp_opt.saw_tstamp &&
231 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
232 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
233 		if (isn == 0)
234 			isn++;
235 		TCP_SKB_CB(skb)->when = isn;
236 		return TCP_TW_SYN;
237 	}
238 
239 	if (paws_reject)
240 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
241 
242 	if (!th->rst) {
243 		/* In this case we must reset the TIMEWAIT timer.
244 		 *
245 		 * If it is ACKless SYN it may be both old duplicate
246 		 * and new good SYN with random sequence number <rcv_nxt.
247 		 * Do not reschedule in the last case.
248 		 */
249 		if (paws_reject || th->ack)
250 			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
251 					   TCP_TIMEWAIT_LEN);
252 
253 		/* Send ACK. Note, we do not put the bucket,
254 		 * it will be released by caller.
255 		 */
256 		return TCP_TW_ACK;
257 	}
258 	inet_twsk_put(tw);
259 	return TCP_TW_SUCCESS;
260 }
261 EXPORT_SYMBOL(tcp_timewait_state_process);
262 
263 /*
264  * Move a socket to time-wait or dead fin-wait-2 state.
265  */
266 void tcp_time_wait(struct sock *sk, int state, int timeo)
267 {
268 	struct inet_timewait_sock *tw = NULL;
269 	const struct inet_connection_sock *icsk = inet_csk(sk);
270 	const struct tcp_sock *tp = tcp_sk(sk);
271 	bool recycle_ok = false;
272 
273 	if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
274 		recycle_ok = tcp_remember_stamp(sk);
275 
276 	if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
277 		tw = inet_twsk_alloc(sk, state);
278 
279 	if (tw != NULL) {
280 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
281 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
282 		struct inet_sock *inet = inet_sk(sk);
283 
284 		tw->tw_transparent	= inet->transparent;
285 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
286 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
287 		tcptw->tw_snd_nxt	= tp->snd_nxt;
288 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
289 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
290 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
291 		tcptw->tw_ts_offset	= tp->tsoffset;
292 
293 #if IS_ENABLED(CONFIG_IPV6)
294 		if (tw->tw_family == PF_INET6) {
295 			struct ipv6_pinfo *np = inet6_sk(sk);
296 
297 			tw->tw_v6_daddr = sk->sk_v6_daddr;
298 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
299 			tw->tw_tclass = np->tclass;
300 			tw->tw_flowlabel = np->flow_label >> 12;
301 			tw->tw_ipv6only = sk->sk_ipv6only;
302 		}
303 #endif
304 
305 #ifdef CONFIG_TCP_MD5SIG
306 		/*
307 		 * The timewait bucket does not have the key DB from the
308 		 * sock structure. We just make a quick copy of the
309 		 * md5 key being used (if indeed we are using one)
310 		 * so the timewait ack generating code has the key.
311 		 */
312 		do {
313 			struct tcp_md5sig_key *key;
314 			tcptw->tw_md5_key = NULL;
315 			key = tp->af_specific->md5_lookup(sk, sk);
316 			if (key != NULL) {
317 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
318 				if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
319 					BUG();
320 			}
321 		} while (0);
322 #endif
323 
324 		/* Linkage updates. */
325 		__inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
326 
327 		/* Get the TIME_WAIT timeout firing. */
328 		if (timeo < rto)
329 			timeo = rto;
330 
331 		if (recycle_ok) {
332 			tw->tw_timeout = rto;
333 		} else {
334 			tw->tw_timeout = TCP_TIMEWAIT_LEN;
335 			if (state == TCP_TIME_WAIT)
336 				timeo = TCP_TIMEWAIT_LEN;
337 		}
338 
339 		inet_twsk_schedule(tw, &tcp_death_row, timeo,
340 				   TCP_TIMEWAIT_LEN);
341 		inet_twsk_put(tw);
342 	} else {
343 		/* Sorry, if we're out of memory, just CLOSE this
344 		 * socket up.  We've got bigger problems than
345 		 * non-graceful socket closings.
346 		 */
347 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
348 	}
349 
350 	tcp_update_metrics(sk);
351 	tcp_done(sk);
352 }
353 
354 void tcp_twsk_destructor(struct sock *sk)
355 {
356 #ifdef CONFIG_TCP_MD5SIG
357 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
358 
359 	if (twsk->tw_md5_key)
360 		kfree_rcu(twsk->tw_md5_key, rcu);
361 #endif
362 }
363 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
364 
365 void tcp_openreq_init_rwin(struct request_sock *req,
366 			   struct sock *sk, struct dst_entry *dst)
367 {
368 	struct inet_request_sock *ireq = inet_rsk(req);
369 	struct tcp_sock *tp = tcp_sk(sk);
370 	__u8 rcv_wscale;
371 	int mss = dst_metric_advmss(dst);
372 
373 	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
374 		mss = tp->rx_opt.user_mss;
375 
376 	/* Set this up on the first call only */
377 	req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
378 
379 	/* limit the window selection if the user enforce a smaller rx buffer */
380 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
381 	    (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0))
382 		req->window_clamp = tcp_full_space(sk);
383 
384 	/* tcp_full_space because it is guaranteed to be the first packet */
385 	tcp_select_initial_window(tcp_full_space(sk),
386 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
387 		&req->rcv_wnd,
388 		&req->window_clamp,
389 		ireq->wscale_ok,
390 		&rcv_wscale,
391 		dst_metric(dst, RTAX_INITRWND));
392 	ireq->rcv_wscale = rcv_wscale;
393 }
394 EXPORT_SYMBOL(tcp_openreq_init_rwin);
395 
396 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
397 					 struct request_sock *req)
398 {
399 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
400 }
401 
402 /* This is not only more efficient than what we used to do, it eliminates
403  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
404  *
405  * Actually, we could lots of memory writes here. tp of listening
406  * socket contains all necessary default parameters.
407  */
408 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
409 {
410 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
411 
412 	if (newsk != NULL) {
413 		const struct inet_request_sock *ireq = inet_rsk(req);
414 		struct tcp_request_sock *treq = tcp_rsk(req);
415 		struct inet_connection_sock *newicsk = inet_csk(newsk);
416 		struct tcp_sock *newtp = tcp_sk(newsk);
417 
418 		/* Now setup tcp_sock */
419 		newtp->pred_flags = 0;
420 
421 		newtp->rcv_wup = newtp->copied_seq =
422 		newtp->rcv_nxt = treq->rcv_isn + 1;
423 
424 		newtp->snd_sml = newtp->snd_una =
425 		newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
426 
427 		tcp_prequeue_init(newtp);
428 		INIT_LIST_HEAD(&newtp->tsq_node);
429 
430 		tcp_init_wl(newtp, treq->rcv_isn);
431 
432 		newtp->srtt_us = 0;
433 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
434 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
435 
436 		newtp->packets_out = 0;
437 		newtp->retrans_out = 0;
438 		newtp->sacked_out = 0;
439 		newtp->fackets_out = 0;
440 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
441 		tcp_enable_early_retrans(newtp);
442 		newtp->tlp_high_seq = 0;
443 		newtp->lsndtime = treq->snt_synack;
444 		newtp->total_retrans = req->num_retrans;
445 
446 		/* So many TCP implementations out there (incorrectly) count the
447 		 * initial SYN frame in their delayed-ACK and congestion control
448 		 * algorithms that we must have the following bandaid to talk
449 		 * efficiently to them.  -DaveM
450 		 */
451 		newtp->snd_cwnd = TCP_INIT_CWND;
452 		newtp->snd_cwnd_cnt = 0;
453 
454 		if (newicsk->icsk_ca_ops != &tcp_init_congestion_ops &&
455 		    !try_module_get(newicsk->icsk_ca_ops->owner))
456 			newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
457 
458 		tcp_set_ca_state(newsk, TCP_CA_Open);
459 		tcp_init_xmit_timers(newsk);
460 		__skb_queue_head_init(&newtp->out_of_order_queue);
461 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
462 
463 		newtp->rx_opt.saw_tstamp = 0;
464 
465 		newtp->rx_opt.dsack = 0;
466 		newtp->rx_opt.num_sacks = 0;
467 
468 		newtp->urg_data = 0;
469 
470 		if (sock_flag(newsk, SOCK_KEEPOPEN))
471 			inet_csk_reset_keepalive_timer(newsk,
472 						       keepalive_time_when(newtp));
473 
474 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
475 		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
476 			if (sysctl_tcp_fack)
477 				tcp_enable_fack(newtp);
478 		}
479 		newtp->window_clamp = req->window_clamp;
480 		newtp->rcv_ssthresh = req->rcv_wnd;
481 		newtp->rcv_wnd = req->rcv_wnd;
482 		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
483 		if (newtp->rx_opt.wscale_ok) {
484 			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
485 			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
486 		} else {
487 			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
488 			newtp->window_clamp = min(newtp->window_clamp, 65535U);
489 		}
490 		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
491 				  newtp->rx_opt.snd_wscale);
492 		newtp->max_window = newtp->snd_wnd;
493 
494 		if (newtp->rx_opt.tstamp_ok) {
495 			newtp->rx_opt.ts_recent = req->ts_recent;
496 			newtp->rx_opt.ts_recent_stamp = get_seconds();
497 			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
498 		} else {
499 			newtp->rx_opt.ts_recent_stamp = 0;
500 			newtp->tcp_header_len = sizeof(struct tcphdr);
501 		}
502 		newtp->tsoffset = 0;
503 #ifdef CONFIG_TCP_MD5SIG
504 		newtp->md5sig_info = NULL;	/*XXX*/
505 		if (newtp->af_specific->md5_lookup(sk, newsk))
506 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
507 #endif
508 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
509 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
510 		newtp->rx_opt.mss_clamp = req->mss;
511 		TCP_ECN_openreq_child(newtp, req);
512 		newtp->fastopen_rsk = NULL;
513 		newtp->syn_data_acked = 0;
514 
515 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
516 	}
517 	return newsk;
518 }
519 EXPORT_SYMBOL(tcp_create_openreq_child);
520 
521 /*
522  * Process an incoming packet for SYN_RECV sockets represented as a
523  * request_sock. Normally sk is the listener socket but for TFO it
524  * points to the child socket.
525  *
526  * XXX (TFO) - The current impl contains a special check for ack
527  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
528  *
529  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
530  */
531 
532 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
533 			   struct request_sock *req,
534 			   struct request_sock **prev,
535 			   bool fastopen)
536 {
537 	struct tcp_options_received tmp_opt;
538 	struct sock *child;
539 	const struct tcphdr *th = tcp_hdr(skb);
540 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
541 	bool paws_reject = false;
542 
543 	BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
544 
545 	tmp_opt.saw_tstamp = 0;
546 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
547 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
548 
549 		if (tmp_opt.saw_tstamp) {
550 			tmp_opt.ts_recent = req->ts_recent;
551 			/* We do not store true stamp, but it is not required,
552 			 * it can be estimated (approximately)
553 			 * from another data.
554 			 */
555 			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
556 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
557 		}
558 	}
559 
560 	/* Check for pure retransmitted SYN. */
561 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
562 	    flg == TCP_FLAG_SYN &&
563 	    !paws_reject) {
564 		/*
565 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
566 		 * this case on figure 6 and figure 8, but formal
567 		 * protocol description says NOTHING.
568 		 * To be more exact, it says that we should send ACK,
569 		 * because this segment (at least, if it has no data)
570 		 * is out of window.
571 		 *
572 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
573 		 *  describe SYN-RECV state. All the description
574 		 *  is wrong, we cannot believe to it and should
575 		 *  rely only on common sense and implementation
576 		 *  experience.
577 		 *
578 		 * Enforce "SYN-ACK" according to figure 8, figure 6
579 		 * of RFC793, fixed by RFC1122.
580 		 *
581 		 * Note that even if there is new data in the SYN packet
582 		 * they will be thrown away too.
583 		 *
584 		 * Reset timer after retransmitting SYNACK, similar to
585 		 * the idea of fast retransmit in recovery.
586 		 */
587 		if (!inet_rtx_syn_ack(sk, req))
588 			req->expires = min(TCP_TIMEOUT_INIT << req->num_timeout,
589 					   TCP_RTO_MAX) + jiffies;
590 		return NULL;
591 	}
592 
593 	/* Further reproduces section "SEGMENT ARRIVES"
594 	   for state SYN-RECEIVED of RFC793.
595 	   It is broken, however, it does not work only
596 	   when SYNs are crossed.
597 
598 	   You would think that SYN crossing is impossible here, since
599 	   we should have a SYN_SENT socket (from connect()) on our end,
600 	   but this is not true if the crossed SYNs were sent to both
601 	   ends by a malicious third party.  We must defend against this,
602 	   and to do that we first verify the ACK (as per RFC793, page
603 	   36) and reset if it is invalid.  Is this a true full defense?
604 	   To convince ourselves, let us consider a way in which the ACK
605 	   test can still pass in this 'malicious crossed SYNs' case.
606 	   Malicious sender sends identical SYNs (and thus identical sequence
607 	   numbers) to both A and B:
608 
609 		A: gets SYN, seq=7
610 		B: gets SYN, seq=7
611 
612 	   By our good fortune, both A and B select the same initial
613 	   send sequence number of seven :-)
614 
615 		A: sends SYN|ACK, seq=7, ack_seq=8
616 		B: sends SYN|ACK, seq=7, ack_seq=8
617 
618 	   So we are now A eating this SYN|ACK, ACK test passes.  So
619 	   does sequence test, SYN is truncated, and thus we consider
620 	   it a bare ACK.
621 
622 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
623 	   bare ACK.  Otherwise, we create an established connection.  Both
624 	   ends (listening sockets) accept the new incoming connection and try
625 	   to talk to each other. 8-)
626 
627 	   Note: This case is both harmless, and rare.  Possibility is about the
628 	   same as us discovering intelligent life on another plant tomorrow.
629 
630 	   But generally, we should (RFC lies!) to accept ACK
631 	   from SYNACK both here and in tcp_rcv_state_process().
632 	   tcp_rcv_state_process() does not, hence, we do not too.
633 
634 	   Note that the case is absolutely generic:
635 	   we cannot optimize anything here without
636 	   violating protocol. All the checks must be made
637 	   before attempt to create socket.
638 	 */
639 
640 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
641 	 *                  and the incoming segment acknowledges something not yet
642 	 *                  sent (the segment carries an unacceptable ACK) ...
643 	 *                  a reset is sent."
644 	 *
645 	 * Invalid ACK: reset will be sent by listening socket.
646 	 * Note that the ACK validity check for a Fast Open socket is done
647 	 * elsewhere and is checked directly against the child socket rather
648 	 * than req because user data may have been sent out.
649 	 */
650 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
651 	    (TCP_SKB_CB(skb)->ack_seq !=
652 	     tcp_rsk(req)->snt_isn + 1))
653 		return sk;
654 
655 	/* Also, it would be not so bad idea to check rcv_tsecr, which
656 	 * is essentially ACK extension and too early or too late values
657 	 * should cause reset in unsynchronized states.
658 	 */
659 
660 	/* RFC793: "first check sequence number". */
661 
662 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
663 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
664 		/* Out of window: send ACK and drop. */
665 		if (!(flg & TCP_FLAG_RST))
666 			req->rsk_ops->send_ack(sk, skb, req);
667 		if (paws_reject)
668 			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
669 		return NULL;
670 	}
671 
672 	/* In sequence, PAWS is OK. */
673 
674 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
675 		req->ts_recent = tmp_opt.rcv_tsval;
676 
677 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
678 		/* Truncate SYN, it is out of window starting
679 		   at tcp_rsk(req)->rcv_isn + 1. */
680 		flg &= ~TCP_FLAG_SYN;
681 	}
682 
683 	/* RFC793: "second check the RST bit" and
684 	 *	   "fourth, check the SYN bit"
685 	 */
686 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
687 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
688 		goto embryonic_reset;
689 	}
690 
691 	/* ACK sequence verified above, just make sure ACK is
692 	 * set.  If ACK not set, just silently drop the packet.
693 	 *
694 	 * XXX (TFO) - if we ever allow "data after SYN", the
695 	 * following check needs to be removed.
696 	 */
697 	if (!(flg & TCP_FLAG_ACK))
698 		return NULL;
699 
700 	/* For Fast Open no more processing is needed (sk is the
701 	 * child socket).
702 	 */
703 	if (fastopen)
704 		return sk;
705 
706 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
707 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
708 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
709 		inet_rsk(req)->acked = 1;
710 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
711 		return NULL;
712 	}
713 
714 	/* OK, ACK is valid, create big socket and
715 	 * feed this segment to it. It will repeat all
716 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
717 	 * ESTABLISHED STATE. If it will be dropped after
718 	 * socket is created, wait for troubles.
719 	 */
720 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
721 	if (child == NULL)
722 		goto listen_overflow;
723 
724 	inet_csk_reqsk_queue_unlink(sk, req, prev);
725 	inet_csk_reqsk_queue_removed(sk, req);
726 
727 	inet_csk_reqsk_queue_add(sk, req, child);
728 	return child;
729 
730 listen_overflow:
731 	if (!sysctl_tcp_abort_on_overflow) {
732 		inet_rsk(req)->acked = 1;
733 		return NULL;
734 	}
735 
736 embryonic_reset:
737 	if (!(flg & TCP_FLAG_RST)) {
738 		/* Received a bad SYN pkt - for TFO We try not to reset
739 		 * the local connection unless it's really necessary to
740 		 * avoid becoming vulnerable to outside attack aiming at
741 		 * resetting legit local connections.
742 		 */
743 		req->rsk_ops->send_reset(sk, skb);
744 	} else if (fastopen) { /* received a valid RST pkt */
745 		reqsk_fastopen_remove(sk, req, true);
746 		tcp_reset(sk);
747 	}
748 	if (!fastopen) {
749 		inet_csk_reqsk_queue_drop(sk, req, prev);
750 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
751 	}
752 	return NULL;
753 }
754 EXPORT_SYMBOL(tcp_check_req);
755 
756 /*
757  * Queue segment on the new socket if the new socket is active,
758  * otherwise we just shortcircuit this and continue with
759  * the new socket.
760  *
761  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
762  * when entering. But other states are possible due to a race condition
763  * where after __inet_lookup_established() fails but before the listener
764  * locked is obtained, other packets cause the same connection to
765  * be created.
766  */
767 
768 int tcp_child_process(struct sock *parent, struct sock *child,
769 		      struct sk_buff *skb)
770 {
771 	int ret = 0;
772 	int state = child->sk_state;
773 
774 	if (!sock_owned_by_user(child)) {
775 		ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
776 					    skb->len);
777 		/* Wakeup parent, send SIGIO */
778 		if (state == TCP_SYN_RECV && child->sk_state != state)
779 			parent->sk_data_ready(parent);
780 	} else {
781 		/* Alas, it is possible again, because we do lookup
782 		 * in main socket hash table and lock on listening
783 		 * socket does not protect us more.
784 		 */
785 		__sk_add_backlog(child, skb);
786 	}
787 
788 	bh_unlock_sock(child);
789 	sock_put(child);
790 	return ret;
791 }
792 EXPORT_SYMBOL(tcp_child_process);
793