xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision 9d749629)
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 	const u8 *hash_location;
97 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
98 	bool paws_reject = false;
99 
100 	tmp_opt.saw_tstamp = 0;
101 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
102 		tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
103 
104 		if (tmp_opt.saw_tstamp) {
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_TW_ACK;
121 
122 		if (th->rst)
123 			goto kill;
124 
125 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
126 			goto kill_with_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 kill_with_rst:
142 			inet_twsk_deschedule(tw, &tcp_death_row);
143 			inet_twsk_put(tw);
144 			return TCP_TW_RST;
145 		}
146 
147 		/* FIN arrived, enter true time-wait state. */
148 		tw->tw_substate	  = TCP_TIME_WAIT;
149 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
150 		if (tmp_opt.saw_tstamp) {
151 			tcptw->tw_ts_recent_stamp = get_seconds();
152 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
153 		}
154 
155 		if (tcp_death_row.sysctl_tw_recycle &&
156 		    tcptw->tw_ts_recent_stamp &&
157 		    tcp_tw_remember_stamp(tw))
158 			inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
159 					   TCP_TIMEWAIT_LEN);
160 		else
161 			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
162 					   TCP_TIMEWAIT_LEN);
163 		return TCP_TW_ACK;
164 	}
165 
166 	/*
167 	 *	Now real TIME-WAIT state.
168 	 *
169 	 *	RFC 1122:
170 	 *	"When a connection is [...] on TIME-WAIT state [...]
171 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
172 	 *	reopen the connection directly, if it:
173 	 *
174 	 *	(1)  assigns its initial sequence number for the new
175 	 *	connection to be larger than the largest sequence
176 	 *	number it used on the previous connection incarnation,
177 	 *	and
178 	 *
179 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
180 	 *	to be an old duplicate".
181 	 */
182 
183 	if (!paws_reject &&
184 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
185 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
186 		/* In window segment, it may be only reset or bare ack. */
187 
188 		if (th->rst) {
189 			/* This is TIME_WAIT assassination, in two flavors.
190 			 * Oh well... nobody has a sufficient solution to this
191 			 * protocol bug yet.
192 			 */
193 			if (sysctl_tcp_rfc1337 == 0) {
194 kill:
195 				inet_twsk_deschedule(tw, &tcp_death_row);
196 				inet_twsk_put(tw);
197 				return TCP_TW_SUCCESS;
198 			}
199 		}
200 		inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
201 				   TCP_TIMEWAIT_LEN);
202 
203 		if (tmp_opt.saw_tstamp) {
204 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
205 			tcptw->tw_ts_recent_stamp = get_seconds();
206 		}
207 
208 		inet_twsk_put(tw);
209 		return TCP_TW_SUCCESS;
210 	}
211 
212 	/* Out of window segment.
213 
214 	   All the segments are ACKed immediately.
215 
216 	   The only exception is new SYN. We accept it, if it is
217 	   not old duplicate and we are not in danger to be killed
218 	   by delayed old duplicates. RFC check is that it has
219 	   newer sequence number works at rates <40Mbit/sec.
220 	   However, if paws works, it is reliable AND even more,
221 	   we even may relax silly seq space cutoff.
222 
223 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
224 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
225 	   we must return socket to time-wait state. It is not good,
226 	   but not fatal yet.
227 	 */
228 
229 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
230 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
231 	     (tmp_opt.saw_tstamp &&
232 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
233 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
234 		if (isn == 0)
235 			isn++;
236 		TCP_SKB_CB(skb)->when = isn;
237 		return TCP_TW_SYN;
238 	}
239 
240 	if (paws_reject)
241 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
242 
243 	if (!th->rst) {
244 		/* In this case we must reset the TIMEWAIT timer.
245 		 *
246 		 * If it is ACKless SYN it may be both old duplicate
247 		 * and new good SYN with random sequence number <rcv_nxt.
248 		 * Do not reschedule in the last case.
249 		 */
250 		if (paws_reject || th->ack)
251 			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
252 					   TCP_TIMEWAIT_LEN);
253 
254 		/* Send ACK. Note, we do not put the bucket,
255 		 * it will be released by caller.
256 		 */
257 		return TCP_TW_ACK;
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 	struct inet_timewait_sock *tw = NULL;
270 	const struct inet_connection_sock *icsk = inet_csk(sk);
271 	const struct tcp_sock *tp = tcp_sk(sk);
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 	if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
278 		tw = inet_twsk_alloc(sk, state);
279 
280 	if (tw != NULL) {
281 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
282 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
283 		struct inet_sock *inet = inet_sk(sk);
284 
285 		tw->tw_transparent	= inet->transparent;
286 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
287 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
288 		tcptw->tw_snd_nxt	= tp->snd_nxt;
289 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
290 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
291 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
292 		tcptw->tw_ts_offset	= tp->tsoffset;
293 
294 #if IS_ENABLED(CONFIG_IPV6)
295 		if (tw->tw_family == PF_INET6) {
296 			struct ipv6_pinfo *np = inet6_sk(sk);
297 			struct inet6_timewait_sock *tw6;
298 
299 			tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
300 			tw6 = inet6_twsk((struct sock *)tw);
301 			tw6->tw_v6_daddr = np->daddr;
302 			tw6->tw_v6_rcv_saddr = np->rcv_saddr;
303 			tw->tw_tclass = np->tclass;
304 			tw->tw_ipv6only = np->ipv6only;
305 		}
306 #endif
307 
308 #ifdef CONFIG_TCP_MD5SIG
309 		/*
310 		 * The timewait bucket does not have the key DB from the
311 		 * sock structure. We just make a quick copy of the
312 		 * md5 key being used (if indeed we are using one)
313 		 * so the timewait ack generating code has the key.
314 		 */
315 		do {
316 			struct tcp_md5sig_key *key;
317 			tcptw->tw_md5_key = NULL;
318 			key = tp->af_specific->md5_lookup(sk, sk);
319 			if (key != NULL) {
320 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
321 				if (tcptw->tw_md5_key && tcp_alloc_md5sig_pool(sk) == NULL)
322 					BUG();
323 			}
324 		} while (0);
325 #endif
326 
327 		/* Linkage updates. */
328 		__inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
329 
330 		/* Get the TIME_WAIT timeout firing. */
331 		if (timeo < rto)
332 			timeo = rto;
333 
334 		if (recycle_ok) {
335 			tw->tw_timeout = rto;
336 		} else {
337 			tw->tw_timeout = TCP_TIMEWAIT_LEN;
338 			if (state == TCP_TIME_WAIT)
339 				timeo = TCP_TIMEWAIT_LEN;
340 		}
341 
342 		inet_twsk_schedule(tw, &tcp_death_row, timeo,
343 				   TCP_TIMEWAIT_LEN);
344 		inet_twsk_put(tw);
345 	} else {
346 		/* Sorry, if we're out of memory, just CLOSE this
347 		 * socket up.  We've got bigger problems than
348 		 * non-graceful socket closings.
349 		 */
350 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
351 	}
352 
353 	tcp_update_metrics(sk);
354 	tcp_done(sk);
355 }
356 
357 void tcp_twsk_destructor(struct sock *sk)
358 {
359 #ifdef CONFIG_TCP_MD5SIG
360 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
361 
362 	if (twsk->tw_md5_key) {
363 		tcp_free_md5sig_pool();
364 		kfree_rcu(twsk->tw_md5_key, rcu);
365 	}
366 #endif
367 }
368 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
369 
370 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
371 					 struct request_sock *req)
372 {
373 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
374 }
375 
376 /* This is not only more efficient than what we used to do, it eliminates
377  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
378  *
379  * Actually, we could lots of memory writes here. tp of listening
380  * socket contains all necessary default parameters.
381  */
382 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
383 {
384 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
385 
386 	if (newsk != NULL) {
387 		const struct inet_request_sock *ireq = inet_rsk(req);
388 		struct tcp_request_sock *treq = tcp_rsk(req);
389 		struct inet_connection_sock *newicsk = inet_csk(newsk);
390 		struct tcp_sock *newtp = tcp_sk(newsk);
391 		struct tcp_sock *oldtp = tcp_sk(sk);
392 		struct tcp_cookie_values *oldcvp = oldtp->cookie_values;
393 
394 		/* TCP Cookie Transactions require space for the cookie pair,
395 		 * as it differs for each connection.  There is no need to
396 		 * copy any s_data_payload stored at the original socket.
397 		 * Failure will prevent resuming the connection.
398 		 *
399 		 * Presumed copied, in order of appearance:
400 		 *	cookie_in_always, cookie_out_never
401 		 */
402 		if (oldcvp != NULL) {
403 			struct tcp_cookie_values *newcvp =
404 				kzalloc(sizeof(*newtp->cookie_values),
405 					GFP_ATOMIC);
406 
407 			if (newcvp != NULL) {
408 				kref_init(&newcvp->kref);
409 				newcvp->cookie_desired =
410 						oldcvp->cookie_desired;
411 				newtp->cookie_values = newcvp;
412 			} else {
413 				/* Not Yet Implemented */
414 				newtp->cookie_values = NULL;
415 			}
416 		}
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 =
426 			treq->snt_isn + 1 + tcp_s_data_size(oldtp);
427 
428 		tcp_prequeue_init(newtp);
429 		INIT_LIST_HEAD(&newtp->tsq_node);
430 
431 		tcp_init_wl(newtp, treq->rcv_isn);
432 
433 		newtp->srtt = 0;
434 		newtp->mdev = TCP_TIMEOUT_INIT;
435 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
436 
437 		newtp->packets_out = 0;
438 		newtp->retrans_out = 0;
439 		newtp->sacked_out = 0;
440 		newtp->fackets_out = 0;
441 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
442 		tcp_enable_early_retrans(newtp);
443 
444 		/* So many TCP implementations out there (incorrectly) count the
445 		 * initial SYN frame in their delayed-ACK and congestion control
446 		 * algorithms that we must have the following bandaid to talk
447 		 * efficiently to them.  -DaveM
448 		 */
449 		newtp->snd_cwnd = TCP_INIT_CWND;
450 		newtp->snd_cwnd_cnt = 0;
451 
452 		newtp->frto_counter = 0;
453 		newtp->frto_highmark = 0;
454 
455 		if (newicsk->icsk_ca_ops != &tcp_init_congestion_ops &&
456 		    !try_module_get(newicsk->icsk_ca_ops->owner))
457 			newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
458 
459 		tcp_set_ca_state(newsk, TCP_CA_Open);
460 		tcp_init_xmit_timers(newsk);
461 		skb_queue_head_init(&newtp->out_of_order_queue);
462 		newtp->write_seq = newtp->pushed_seq =
463 			treq->snt_isn + 1 + tcp_s_data_size(oldtp);
464 
465 		newtp->rx_opt.saw_tstamp = 0;
466 
467 		newtp->rx_opt.dsack = 0;
468 		newtp->rx_opt.num_sacks = 0;
469 
470 		newtp->urg_data = 0;
471 
472 		if (sock_flag(newsk, SOCK_KEEPOPEN))
473 			inet_csk_reset_keepalive_timer(newsk,
474 						       keepalive_time_when(newtp));
475 
476 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
477 		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
478 			if (sysctl_tcp_fack)
479 				tcp_enable_fack(newtp);
480 		}
481 		newtp->window_clamp = req->window_clamp;
482 		newtp->rcv_ssthresh = req->rcv_wnd;
483 		newtp->rcv_wnd = req->rcv_wnd;
484 		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
485 		if (newtp->rx_opt.wscale_ok) {
486 			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
487 			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
488 		} else {
489 			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
490 			newtp->window_clamp = min(newtp->window_clamp, 65535U);
491 		}
492 		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
493 				  newtp->rx_opt.snd_wscale);
494 		newtp->max_window = newtp->snd_wnd;
495 
496 		if (newtp->rx_opt.tstamp_ok) {
497 			newtp->rx_opt.ts_recent = req->ts_recent;
498 			newtp->rx_opt.ts_recent_stamp = get_seconds();
499 			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
500 		} else {
501 			newtp->rx_opt.ts_recent_stamp = 0;
502 			newtp->tcp_header_len = sizeof(struct tcphdr);
503 		}
504 		newtp->tsoffset = 0;
505 #ifdef CONFIG_TCP_MD5SIG
506 		newtp->md5sig_info = NULL;	/*XXX*/
507 		if (newtp->af_specific->md5_lookup(sk, newsk))
508 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
509 #endif
510 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
511 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
512 		newtp->rx_opt.mss_clamp = req->mss;
513 		TCP_ECN_openreq_child(newtp, req);
514 		newtp->fastopen_rsk = NULL;
515 		newtp->syn_data_acked = 0;
516 
517 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
518 	}
519 	return newsk;
520 }
521 EXPORT_SYMBOL(tcp_create_openreq_child);
522 
523 /*
524  * Process an incoming packet for SYN_RECV sockets represented as a
525  * request_sock. Normally sk is the listener socket but for TFO it
526  * points to the child socket.
527  *
528  * XXX (TFO) - The current impl contains a special check for ack
529  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
530  *
531  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
532  */
533 
534 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
535 			   struct request_sock *req,
536 			   struct request_sock **prev,
537 			   bool fastopen)
538 {
539 	struct tcp_options_received tmp_opt;
540 	const u8 *hash_location;
541 	struct sock *child;
542 	const struct tcphdr *th = tcp_hdr(skb);
543 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
544 	bool paws_reject = false;
545 
546 	BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
547 
548 	tmp_opt.saw_tstamp = 0;
549 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
550 		tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
551 
552 		if (tmp_opt.saw_tstamp) {
553 			tmp_opt.ts_recent = req->ts_recent;
554 			/* We do not store true stamp, but it is not required,
555 			 * it can be estimated (approximately)
556 			 * from another data.
557 			 */
558 			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
559 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
560 		}
561 	}
562 
563 	/* Check for pure retransmitted SYN. */
564 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
565 	    flg == TCP_FLAG_SYN &&
566 	    !paws_reject) {
567 		/*
568 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
569 		 * this case on figure 6 and figure 8, but formal
570 		 * protocol description says NOTHING.
571 		 * To be more exact, it says that we should send ACK,
572 		 * because this segment (at least, if it has no data)
573 		 * is out of window.
574 		 *
575 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
576 		 *  describe SYN-RECV state. All the description
577 		 *  is wrong, we cannot believe to it and should
578 		 *  rely only on common sense and implementation
579 		 *  experience.
580 		 *
581 		 * Enforce "SYN-ACK" according to figure 8, figure 6
582 		 * of RFC793, fixed by RFC1122.
583 		 *
584 		 * Note that even if there is new data in the SYN packet
585 		 * they will be thrown away too.
586 		 */
587 		inet_rtx_syn_ack(sk, req);
588 		return NULL;
589 	}
590 
591 	/* Further reproduces section "SEGMENT ARRIVES"
592 	   for state SYN-RECEIVED of RFC793.
593 	   It is broken, however, it does not work only
594 	   when SYNs are crossed.
595 
596 	   You would think that SYN crossing is impossible here, since
597 	   we should have a SYN_SENT socket (from connect()) on our end,
598 	   but this is not true if the crossed SYNs were sent to both
599 	   ends by a malicious third party.  We must defend against this,
600 	   and to do that we first verify the ACK (as per RFC793, page
601 	   36) and reset if it is invalid.  Is this a true full defense?
602 	   To convince ourselves, let us consider a way in which the ACK
603 	   test can still pass in this 'malicious crossed SYNs' case.
604 	   Malicious sender sends identical SYNs (and thus identical sequence
605 	   numbers) to both A and B:
606 
607 		A: gets SYN, seq=7
608 		B: gets SYN, seq=7
609 
610 	   By our good fortune, both A and B select the same initial
611 	   send sequence number of seven :-)
612 
613 		A: sends SYN|ACK, seq=7, ack_seq=8
614 		B: sends SYN|ACK, seq=7, ack_seq=8
615 
616 	   So we are now A eating this SYN|ACK, ACK test passes.  So
617 	   does sequence test, SYN is truncated, and thus we consider
618 	   it a bare ACK.
619 
620 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
621 	   bare ACK.  Otherwise, we create an established connection.  Both
622 	   ends (listening sockets) accept the new incoming connection and try
623 	   to talk to each other. 8-)
624 
625 	   Note: This case is both harmless, and rare.  Possibility is about the
626 	   same as us discovering intelligent life on another plant tomorrow.
627 
628 	   But generally, we should (RFC lies!) to accept ACK
629 	   from SYNACK both here and in tcp_rcv_state_process().
630 	   tcp_rcv_state_process() does not, hence, we do not too.
631 
632 	   Note that the case is absolutely generic:
633 	   we cannot optimize anything here without
634 	   violating protocol. All the checks must be made
635 	   before attempt to create socket.
636 	 */
637 
638 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
639 	 *                  and the incoming segment acknowledges something not yet
640 	 *                  sent (the segment carries an unacceptable ACK) ...
641 	 *                  a reset is sent."
642 	 *
643 	 * Invalid ACK: reset will be sent by listening socket.
644 	 * Note that the ACK validity check for a Fast Open socket is done
645 	 * elsewhere and is checked directly against the child socket rather
646 	 * than req because user data may have been sent out.
647 	 */
648 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
649 	    (TCP_SKB_CB(skb)->ack_seq !=
650 	     tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk))))
651 		return sk;
652 
653 	/* Also, it would be not so bad idea to check rcv_tsecr, which
654 	 * is essentially ACK extension and too early or too late values
655 	 * should cause reset in unsynchronized states.
656 	 */
657 
658 	/* RFC793: "first check sequence number". */
659 
660 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
661 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
662 		/* Out of window: send ACK and drop. */
663 		if (!(flg & TCP_FLAG_RST))
664 			req->rsk_ops->send_ack(sk, skb, req);
665 		if (paws_reject)
666 			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
667 		return NULL;
668 	}
669 
670 	/* In sequence, PAWS is OK. */
671 
672 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
673 		req->ts_recent = tmp_opt.rcv_tsval;
674 
675 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
676 		/* Truncate SYN, it is out of window starting
677 		   at tcp_rsk(req)->rcv_isn + 1. */
678 		flg &= ~TCP_FLAG_SYN;
679 	}
680 
681 	/* RFC793: "second check the RST bit" and
682 	 *	   "fourth, check the SYN bit"
683 	 */
684 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
685 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
686 		goto embryonic_reset;
687 	}
688 
689 	/* ACK sequence verified above, just make sure ACK is
690 	 * set.  If ACK not set, just silently drop the packet.
691 	 *
692 	 * XXX (TFO) - if we ever allow "data after SYN", the
693 	 * following check needs to be removed.
694 	 */
695 	if (!(flg & TCP_FLAG_ACK))
696 		return NULL;
697 
698 	/* Got ACK for our SYNACK, so update baseline for SYNACK RTT sample. */
699 	if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
700 		tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
701 	else if (req->num_retrans) /* don't take RTT sample if retrans && ~TS */
702 		tcp_rsk(req)->snt_synack = 0;
703 
704 	/* For Fast Open no more processing is needed (sk is the
705 	 * child socket).
706 	 */
707 	if (fastopen)
708 		return sk;
709 
710 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
711 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
712 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
713 		inet_rsk(req)->acked = 1;
714 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
715 		return NULL;
716 	}
717 
718 	/* OK, ACK is valid, create big socket and
719 	 * feed this segment to it. It will repeat all
720 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
721 	 * ESTABLISHED STATE. If it will be dropped after
722 	 * socket is created, wait for troubles.
723 	 */
724 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
725 	if (child == NULL)
726 		goto listen_overflow;
727 
728 	inet_csk_reqsk_queue_unlink(sk, req, prev);
729 	inet_csk_reqsk_queue_removed(sk, req);
730 
731 	inet_csk_reqsk_queue_add(sk, req, child);
732 	return child;
733 
734 listen_overflow:
735 	if (!sysctl_tcp_abort_on_overflow) {
736 		inet_rsk(req)->acked = 1;
737 		return NULL;
738 	}
739 
740 embryonic_reset:
741 	if (!(flg & TCP_FLAG_RST)) {
742 		/* Received a bad SYN pkt - for TFO We try not to reset
743 		 * the local connection unless it's really necessary to
744 		 * avoid becoming vulnerable to outside attack aiming at
745 		 * resetting legit local connections.
746 		 */
747 		req->rsk_ops->send_reset(sk, skb);
748 	} else if (fastopen) { /* received a valid RST pkt */
749 		reqsk_fastopen_remove(sk, req, true);
750 		tcp_reset(sk);
751 	}
752 	if (!fastopen) {
753 		inet_csk_reqsk_queue_drop(sk, req, prev);
754 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
755 	}
756 	return NULL;
757 }
758 EXPORT_SYMBOL(tcp_check_req);
759 
760 /*
761  * Queue segment on the new socket if the new socket is active,
762  * otherwise we just shortcircuit this and continue with
763  * the new socket.
764  *
765  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
766  * when entering. But other states are possible due to a race condition
767  * where after __inet_lookup_established() fails but before the listener
768  * locked is obtained, other packets cause the same connection to
769  * be created.
770  */
771 
772 int tcp_child_process(struct sock *parent, struct sock *child,
773 		      struct sk_buff *skb)
774 {
775 	int ret = 0;
776 	int state = child->sk_state;
777 
778 	if (!sock_owned_by_user(child)) {
779 		ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
780 					    skb->len);
781 		/* Wakeup parent, send SIGIO */
782 		if (state == TCP_SYN_RECV && child->sk_state != state)
783 			parent->sk_data_ready(parent, 0);
784 	} else {
785 		/* Alas, it is possible again, because we do lookup
786 		 * in main socket hash table and lock on listening
787 		 * socket does not protect us more.
788 		 */
789 		__sk_add_backlog(child, skb);
790 	}
791 
792 	bh_unlock_sock(child);
793 	sock_put(child);
794 	return ret;
795 }
796 EXPORT_SYMBOL(tcp_child_process);
797