xref: /linux/net/ipv4/tcp_minisocks.c (revision 63307d015b91e626c97bb82e88054af3d0b74643)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Implementation of the Transmission Control Protocol(TCP).
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
12  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
13  *		Florian La Roche, <flla@stud.uni-sb.de>
14  *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
15  *		Linus Torvalds, <torvalds@cs.helsinki.fi>
16  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
17  *		Matthew Dillon, <dillon@apollo.west.oic.com>
18  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
19  *		Jorge Cwik, <jorge@laser.satlink.net>
20  */
21 
22 #include <linux/mm.h>
23 #include <linux/module.h>
24 #include <linux/slab.h>
25 #include <linux/sysctl.h>
26 #include <linux/workqueue.h>
27 #include <linux/static_key.h>
28 #include <net/tcp.h>
29 #include <net/inet_common.h>
30 #include <net/xfrm.h>
31 #include <net/busy_poll.h>
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(twsk_net(tw), 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 = ktime_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 (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) {
184 kill:
185 				inet_twsk_deschedule_put(tw);
186 				return TCP_TW_SUCCESS;
187 			}
188 		} else {
189 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
190 		}
191 
192 		if (tmp_opt.saw_tstamp) {
193 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
194 			tcptw->tw_ts_recent_stamp = ktime_get_seconds();
195 		}
196 
197 		inet_twsk_put(tw);
198 		return TCP_TW_SUCCESS;
199 	}
200 
201 	/* Out of window segment.
202 
203 	   All the segments are ACKed immediately.
204 
205 	   The only exception is new SYN. We accept it, if it is
206 	   not old duplicate and we are not in danger to be killed
207 	   by delayed old duplicates. RFC check is that it has
208 	   newer sequence number works at rates <40Mbit/sec.
209 	   However, if paws works, it is reliable AND even more,
210 	   we even may relax silly seq space cutoff.
211 
212 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
213 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
214 	   we must return socket to time-wait state. It is not good,
215 	   but not fatal yet.
216 	 */
217 
218 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
219 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
220 	     (tmp_opt.saw_tstamp &&
221 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
222 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
223 		if (isn == 0)
224 			isn++;
225 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
226 		return TCP_TW_SYN;
227 	}
228 
229 	if (paws_reject)
230 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
231 
232 	if (!th->rst) {
233 		/* In this case we must reset the TIMEWAIT timer.
234 		 *
235 		 * If it is ACKless SYN it may be both old duplicate
236 		 * and new good SYN with random sequence number <rcv_nxt.
237 		 * Do not reschedule in the last case.
238 		 */
239 		if (paws_reject || th->ack)
240 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
241 
242 		return tcp_timewait_check_oow_rate_limit(
243 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
244 	}
245 	inet_twsk_put(tw);
246 	return TCP_TW_SUCCESS;
247 }
248 EXPORT_SYMBOL(tcp_timewait_state_process);
249 
250 /*
251  * Move a socket to time-wait or dead fin-wait-2 state.
252  */
253 void tcp_time_wait(struct sock *sk, int state, int timeo)
254 {
255 	const struct inet_connection_sock *icsk = inet_csk(sk);
256 	const struct tcp_sock *tp = tcp_sk(sk);
257 	struct inet_timewait_sock *tw;
258 	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
259 
260 	tw = inet_twsk_alloc(sk, tcp_death_row, state);
261 
262 	if (tw) {
263 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
264 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
265 		struct inet_sock *inet = inet_sk(sk);
266 
267 		tw->tw_transparent	= inet->transparent;
268 		tw->tw_mark		= sk->sk_mark;
269 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
270 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
271 		tcptw->tw_snd_nxt	= tp->snd_nxt;
272 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
273 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
274 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
275 		tcptw->tw_ts_offset	= tp->tsoffset;
276 		tcptw->tw_last_oow_ack_time = 0;
277 
278 #if IS_ENABLED(CONFIG_IPV6)
279 		if (tw->tw_family == PF_INET6) {
280 			struct ipv6_pinfo *np = inet6_sk(sk);
281 
282 			tw->tw_v6_daddr = sk->sk_v6_daddr;
283 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
284 			tw->tw_tclass = np->tclass;
285 			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
286 			tw->tw_ipv6only = sk->sk_ipv6only;
287 		}
288 #endif
289 
290 #ifdef CONFIG_TCP_MD5SIG
291 		/*
292 		 * The timewait bucket does not have the key DB from the
293 		 * sock structure. We just make a quick copy of the
294 		 * md5 key being used (if indeed we are using one)
295 		 * so the timewait ack generating code has the key.
296 		 */
297 		do {
298 			tcptw->tw_md5_key = NULL;
299 			if (static_branch_unlikely(&tcp_md5_needed)) {
300 				struct tcp_md5sig_key *key;
301 
302 				key = tp->af_specific->md5_lookup(sk, sk);
303 				if (key) {
304 					tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
305 					BUG_ON(tcptw->tw_md5_key && !tcp_alloc_md5sig_pool());
306 				}
307 			}
308 		} while (0);
309 #endif
310 
311 		/* Get the TIME_WAIT timeout firing. */
312 		if (timeo < rto)
313 			timeo = rto;
314 
315 		if (state == TCP_TIME_WAIT)
316 			timeo = TCP_TIMEWAIT_LEN;
317 
318 		/* tw_timer is pinned, so we need to make sure BH are disabled
319 		 * in following section, otherwise timer handler could run before
320 		 * we complete the initialization.
321 		 */
322 		local_bh_disable();
323 		inet_twsk_schedule(tw, timeo);
324 		/* Linkage updates.
325 		 * Note that access to tw after this point is illegal.
326 		 */
327 		inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
328 		local_bh_enable();
329 	} else {
330 		/* Sorry, if we're out of memory, just CLOSE this
331 		 * socket up.  We've got bigger problems than
332 		 * non-graceful socket closings.
333 		 */
334 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
335 	}
336 
337 	tcp_update_metrics(sk);
338 	tcp_done(sk);
339 }
340 EXPORT_SYMBOL(tcp_time_wait);
341 
342 void tcp_twsk_destructor(struct sock *sk)
343 {
344 #ifdef CONFIG_TCP_MD5SIG
345 	if (static_branch_unlikely(&tcp_md5_needed)) {
346 		struct tcp_timewait_sock *twsk = tcp_twsk(sk);
347 
348 		if (twsk->tw_md5_key)
349 			kfree_rcu(twsk->tw_md5_key, rcu);
350 	}
351 #endif
352 }
353 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
354 
355 /* Warning : This function is called without sk_listener being locked.
356  * Be sure to read socket fields once, as their value could change under us.
357  */
358 void tcp_openreq_init_rwin(struct request_sock *req,
359 			   const struct sock *sk_listener,
360 			   const struct dst_entry *dst)
361 {
362 	struct inet_request_sock *ireq = inet_rsk(req);
363 	const struct tcp_sock *tp = tcp_sk(sk_listener);
364 	int full_space = tcp_full_space(sk_listener);
365 	u32 window_clamp;
366 	__u8 rcv_wscale;
367 	u32 rcv_wnd;
368 	int mss;
369 
370 	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
371 	window_clamp = READ_ONCE(tp->window_clamp);
372 	/* Set this up on the first call only */
373 	req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
374 
375 	/* limit the window selection if the user enforce a smaller rx buffer */
376 	if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
377 	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
378 		req->rsk_window_clamp = full_space;
379 
380 	rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
381 	if (rcv_wnd == 0)
382 		rcv_wnd = dst_metric(dst, RTAX_INITRWND);
383 	else if (full_space < rcv_wnd * mss)
384 		full_space = rcv_wnd * mss;
385 
386 	/* tcp_full_space because it is guaranteed to be the first packet */
387 	tcp_select_initial_window(sk_listener, full_space,
388 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
389 		&req->rsk_rcv_wnd,
390 		&req->rsk_window_clamp,
391 		ireq->wscale_ok,
392 		&rcv_wscale,
393 		rcv_wnd);
394 	ireq->rcv_wscale = rcv_wscale;
395 }
396 EXPORT_SYMBOL(tcp_openreq_init_rwin);
397 
398 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
399 				  const struct request_sock *req)
400 {
401 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
402 }
403 
404 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
405 {
406 	struct inet_connection_sock *icsk = inet_csk(sk);
407 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
408 	bool ca_got_dst = false;
409 
410 	if (ca_key != TCP_CA_UNSPEC) {
411 		const struct tcp_congestion_ops *ca;
412 
413 		rcu_read_lock();
414 		ca = tcp_ca_find_key(ca_key);
415 		if (likely(ca && try_module_get(ca->owner))) {
416 			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
417 			icsk->icsk_ca_ops = ca;
418 			ca_got_dst = true;
419 		}
420 		rcu_read_unlock();
421 	}
422 
423 	/* If no valid choice made yet, assign current system default ca. */
424 	if (!ca_got_dst &&
425 	    (!icsk->icsk_ca_setsockopt ||
426 	     !try_module_get(icsk->icsk_ca_ops->owner)))
427 		tcp_assign_congestion_control(sk);
428 
429 	tcp_set_ca_state(sk, TCP_CA_Open);
430 }
431 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
432 
433 static void smc_check_reset_syn_req(struct tcp_sock *oldtp,
434 				    struct request_sock *req,
435 				    struct tcp_sock *newtp)
436 {
437 #if IS_ENABLED(CONFIG_SMC)
438 	struct inet_request_sock *ireq;
439 
440 	if (static_branch_unlikely(&tcp_have_smc)) {
441 		ireq = inet_rsk(req);
442 		if (oldtp->syn_smc && !ireq->smc_ok)
443 			newtp->syn_smc = 0;
444 	}
445 #endif
446 }
447 
448 /* This is not only more efficient than what we used to do, it eliminates
449  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
450  *
451  * Actually, we could lots of memory writes here. tp of listening
452  * socket contains all necessary default parameters.
453  */
454 struct sock *tcp_create_openreq_child(const struct sock *sk,
455 				      struct request_sock *req,
456 				      struct sk_buff *skb)
457 {
458 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
459 	const struct inet_request_sock *ireq = inet_rsk(req);
460 	struct tcp_request_sock *treq = tcp_rsk(req);
461 	struct inet_connection_sock *newicsk;
462 	struct tcp_sock *oldtp, *newtp;
463 
464 	if (!newsk)
465 		return NULL;
466 
467 	newicsk = inet_csk(newsk);
468 	newtp = tcp_sk(newsk);
469 	oldtp = tcp_sk(sk);
470 
471 	smc_check_reset_syn_req(oldtp, req, newtp);
472 
473 	/* Now setup tcp_sock */
474 	newtp->pred_flags = 0;
475 
476 	newtp->rcv_wup = newtp->copied_seq =
477 	newtp->rcv_nxt = treq->rcv_isn + 1;
478 	newtp->segs_in = 1;
479 
480 	newtp->snd_sml = newtp->snd_una =
481 	newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
482 
483 	INIT_LIST_HEAD(&newtp->tsq_node);
484 	INIT_LIST_HEAD(&newtp->tsorted_sent_queue);
485 
486 	tcp_init_wl(newtp, treq->rcv_isn);
487 
488 	minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
489 	newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
490 
491 	newtp->lsndtime = tcp_jiffies32;
492 	newsk->sk_txhash = treq->txhash;
493 	newtp->total_retrans = req->num_retrans;
494 
495 	tcp_init_xmit_timers(newsk);
496 	newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
497 
498 	if (sock_flag(newsk, SOCK_KEEPOPEN))
499 		inet_csk_reset_keepalive_timer(newsk,
500 					       keepalive_time_when(newtp));
501 
502 	newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
503 	newtp->rx_opt.sack_ok = ireq->sack_ok;
504 	newtp->window_clamp = req->rsk_window_clamp;
505 	newtp->rcv_ssthresh = req->rsk_rcv_wnd;
506 	newtp->rcv_wnd = req->rsk_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) << newtp->rx_opt.snd_wscale;
516 	newtp->max_window = newtp->snd_wnd;
517 
518 	if (newtp->rx_opt.tstamp_ok) {
519 		newtp->rx_opt.ts_recent = req->ts_recent;
520 		newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
521 		newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
522 	} else {
523 		newtp->rx_opt.ts_recent_stamp = 0;
524 		newtp->tcp_header_len = sizeof(struct tcphdr);
525 	}
526 	if (req->num_timeout) {
527 		newtp->undo_marker = treq->snt_isn;
528 		newtp->retrans_stamp = div_u64(treq->snt_synack,
529 					       USEC_PER_SEC / TCP_TS_HZ);
530 	}
531 	newtp->tsoffset = treq->ts_off;
532 #ifdef CONFIG_TCP_MD5SIG
533 	newtp->md5sig_info = NULL;	/*XXX*/
534 	if (newtp->af_specific->md5_lookup(sk, newsk))
535 		newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
536 #endif
537 	if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
538 		newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
539 	newtp->rx_opt.mss_clamp = req->mss;
540 	tcp_ecn_openreq_child(newtp, req);
541 	newtp->fastopen_req = NULL;
542 	newtp->fastopen_rsk = NULL;
543 
544 	__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
545 
546 	return newsk;
547 }
548 EXPORT_SYMBOL(tcp_create_openreq_child);
549 
550 /*
551  * Process an incoming packet for SYN_RECV sockets represented as a
552  * request_sock. Normally sk is the listener socket but for TFO it
553  * points to the child socket.
554  *
555  * XXX (TFO) - The current impl contains a special check for ack
556  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
557  *
558  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
559  */
560 
561 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
562 			   struct request_sock *req,
563 			   bool fastopen, bool *req_stolen)
564 {
565 	struct tcp_options_received tmp_opt;
566 	struct sock *child;
567 	const struct tcphdr *th = tcp_hdr(skb);
568 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
569 	bool paws_reject = false;
570 	bool own_req;
571 
572 	tmp_opt.saw_tstamp = 0;
573 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
574 		tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
575 
576 		if (tmp_opt.saw_tstamp) {
577 			tmp_opt.ts_recent = req->ts_recent;
578 			if (tmp_opt.rcv_tsecr)
579 				tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
580 			/* We do not store true stamp, but it is not required,
581 			 * it can be estimated (approximately)
582 			 * from another data.
583 			 */
584 			tmp_opt.ts_recent_stamp = ktime_get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
585 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
586 		}
587 	}
588 
589 	/* Check for pure retransmitted SYN. */
590 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
591 	    flg == TCP_FLAG_SYN &&
592 	    !paws_reject) {
593 		/*
594 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
595 		 * this case on figure 6 and figure 8, but formal
596 		 * protocol description says NOTHING.
597 		 * To be more exact, it says that we should send ACK,
598 		 * because this segment (at least, if it has no data)
599 		 * is out of window.
600 		 *
601 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
602 		 *  describe SYN-RECV state. All the description
603 		 *  is wrong, we cannot believe to it and should
604 		 *  rely only on common sense and implementation
605 		 *  experience.
606 		 *
607 		 * Enforce "SYN-ACK" according to figure 8, figure 6
608 		 * of RFC793, fixed by RFC1122.
609 		 *
610 		 * Note that even if there is new data in the SYN packet
611 		 * they will be thrown away too.
612 		 *
613 		 * Reset timer after retransmitting SYNACK, similar to
614 		 * the idea of fast retransmit in recovery.
615 		 */
616 		if (!tcp_oow_rate_limited(sock_net(sk), skb,
617 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
618 					  &tcp_rsk(req)->last_oow_ack_time) &&
619 
620 		    !inet_rtx_syn_ack(sk, req)) {
621 			unsigned long expires = jiffies;
622 
623 			expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
624 				       TCP_RTO_MAX);
625 			if (!fastopen)
626 				mod_timer_pending(&req->rsk_timer, expires);
627 			else
628 				req->rsk_timer.expires = expires;
629 		}
630 		return NULL;
631 	}
632 
633 	/* Further reproduces section "SEGMENT ARRIVES"
634 	   for state SYN-RECEIVED of RFC793.
635 	   It is broken, however, it does not work only
636 	   when SYNs are crossed.
637 
638 	   You would think that SYN crossing is impossible here, since
639 	   we should have a SYN_SENT socket (from connect()) on our end,
640 	   but this is not true if the crossed SYNs were sent to both
641 	   ends by a malicious third party.  We must defend against this,
642 	   and to do that we first verify the ACK (as per RFC793, page
643 	   36) and reset if it is invalid.  Is this a true full defense?
644 	   To convince ourselves, let us consider a way in which the ACK
645 	   test can still pass in this 'malicious crossed SYNs' case.
646 	   Malicious sender sends identical SYNs (and thus identical sequence
647 	   numbers) to both A and B:
648 
649 		A: gets SYN, seq=7
650 		B: gets SYN, seq=7
651 
652 	   By our good fortune, both A and B select the same initial
653 	   send sequence number of seven :-)
654 
655 		A: sends SYN|ACK, seq=7, ack_seq=8
656 		B: sends SYN|ACK, seq=7, ack_seq=8
657 
658 	   So we are now A eating this SYN|ACK, ACK test passes.  So
659 	   does sequence test, SYN is truncated, and thus we consider
660 	   it a bare ACK.
661 
662 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
663 	   bare ACK.  Otherwise, we create an established connection.  Both
664 	   ends (listening sockets) accept the new incoming connection and try
665 	   to talk to each other. 8-)
666 
667 	   Note: This case is both harmless, and rare.  Possibility is about the
668 	   same as us discovering intelligent life on another plant tomorrow.
669 
670 	   But generally, we should (RFC lies!) to accept ACK
671 	   from SYNACK both here and in tcp_rcv_state_process().
672 	   tcp_rcv_state_process() does not, hence, we do not too.
673 
674 	   Note that the case is absolutely generic:
675 	   we cannot optimize anything here without
676 	   violating protocol. All the checks must be made
677 	   before attempt to create socket.
678 	 */
679 
680 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
681 	 *                  and the incoming segment acknowledges something not yet
682 	 *                  sent (the segment carries an unacceptable ACK) ...
683 	 *                  a reset is sent."
684 	 *
685 	 * Invalid ACK: reset will be sent by listening socket.
686 	 * Note that the ACK validity check for a Fast Open socket is done
687 	 * elsewhere and is checked directly against the child socket rather
688 	 * than req because user data may have been sent out.
689 	 */
690 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
691 	    (TCP_SKB_CB(skb)->ack_seq !=
692 	     tcp_rsk(req)->snt_isn + 1))
693 		return sk;
694 
695 	/* Also, it would be not so bad idea to check rcv_tsecr, which
696 	 * is essentially ACK extension and too early or too late values
697 	 * should cause reset in unsynchronized states.
698 	 */
699 
700 	/* RFC793: "first check sequence number". */
701 
702 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
703 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
704 		/* Out of window: send ACK and drop. */
705 		if (!(flg & TCP_FLAG_RST) &&
706 		    !tcp_oow_rate_limited(sock_net(sk), skb,
707 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
708 					  &tcp_rsk(req)->last_oow_ack_time))
709 			req->rsk_ops->send_ack(sk, skb, req);
710 		if (paws_reject)
711 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
712 		return NULL;
713 	}
714 
715 	/* In sequence, PAWS is OK. */
716 
717 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
718 		req->ts_recent = tmp_opt.rcv_tsval;
719 
720 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
721 		/* Truncate SYN, it is out of window starting
722 		   at tcp_rsk(req)->rcv_isn + 1. */
723 		flg &= ~TCP_FLAG_SYN;
724 	}
725 
726 	/* RFC793: "second check the RST bit" and
727 	 *	   "fourth, check the SYN bit"
728 	 */
729 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
730 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
731 		goto embryonic_reset;
732 	}
733 
734 	/* ACK sequence verified above, just make sure ACK is
735 	 * set.  If ACK not set, just silently drop the packet.
736 	 *
737 	 * XXX (TFO) - if we ever allow "data after SYN", the
738 	 * following check needs to be removed.
739 	 */
740 	if (!(flg & TCP_FLAG_ACK))
741 		return NULL;
742 
743 	/* For Fast Open no more processing is needed (sk is the
744 	 * child socket).
745 	 */
746 	if (fastopen)
747 		return sk;
748 
749 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
750 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
751 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
752 		inet_rsk(req)->acked = 1;
753 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
754 		return NULL;
755 	}
756 
757 	/* OK, ACK is valid, create big socket and
758 	 * feed this segment to it. It will repeat all
759 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
760 	 * ESTABLISHED STATE. If it will be dropped after
761 	 * socket is created, wait for troubles.
762 	 */
763 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
764 							 req, &own_req);
765 	if (!child)
766 		goto listen_overflow;
767 
768 	sock_rps_save_rxhash(child, skb);
769 	tcp_synack_rtt_meas(child, req);
770 	*req_stolen = !own_req;
771 	return inet_csk_complete_hashdance(sk, child, req, own_req);
772 
773 listen_overflow:
774 	if (!sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow) {
775 		inet_rsk(req)->acked = 1;
776 		return NULL;
777 	}
778 
779 embryonic_reset:
780 	if (!(flg & TCP_FLAG_RST)) {
781 		/* Received a bad SYN pkt - for TFO We try not to reset
782 		 * the local connection unless it's really necessary to
783 		 * avoid becoming vulnerable to outside attack aiming at
784 		 * resetting legit local connections.
785 		 */
786 		req->rsk_ops->send_reset(sk, skb);
787 	} else if (fastopen) { /* received a valid RST pkt */
788 		reqsk_fastopen_remove(sk, req, true);
789 		tcp_reset(sk);
790 	}
791 	if (!fastopen) {
792 		inet_csk_reqsk_queue_drop(sk, req);
793 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
794 	}
795 	return NULL;
796 }
797 EXPORT_SYMBOL(tcp_check_req);
798 
799 /*
800  * Queue segment on the new socket if the new socket is active,
801  * otherwise we just shortcircuit this and continue with
802  * the new socket.
803  *
804  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
805  * when entering. But other states are possible due to a race condition
806  * where after __inet_lookup_established() fails but before the listener
807  * locked is obtained, other packets cause the same connection to
808  * be created.
809  */
810 
811 int tcp_child_process(struct sock *parent, struct sock *child,
812 		      struct sk_buff *skb)
813 {
814 	int ret = 0;
815 	int state = child->sk_state;
816 
817 	/* record NAPI ID of child */
818 	sk_mark_napi_id(child, skb);
819 
820 	tcp_segs_in(tcp_sk(child), skb);
821 	if (!sock_owned_by_user(child)) {
822 		ret = tcp_rcv_state_process(child, skb);
823 		/* Wakeup parent, send SIGIO */
824 		if (state == TCP_SYN_RECV && child->sk_state != state)
825 			parent->sk_data_ready(parent);
826 	} else {
827 		/* Alas, it is possible again, because we do lookup
828 		 * in main socket hash table and lock on listening
829 		 * socket does not protect us more.
830 		 */
831 		__sk_add_backlog(child, skb);
832 	}
833 
834 	bh_unlock_sock(child);
835 	sock_put(child);
836 	return ret;
837 }
838 EXPORT_SYMBOL(tcp_child_process);
839