xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision be709d48)
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 <linux/static_key.h>
27 #include <net/tcp.h>
28 #include <net/inet_common.h>
29 #include <net/xfrm.h>
30 #include <net/busy_poll.h>
31 
32 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
33 {
34 	if (seq == s_win)
35 		return true;
36 	if (after(end_seq, s_win) && before(seq, e_win))
37 		return true;
38 	return seq == e_win && seq == end_seq;
39 }
40 
41 static enum tcp_tw_status
42 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
43 				  const struct sk_buff *skb, int mib_idx)
44 {
45 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
46 
47 	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
48 				  &tcptw->tw_last_oow_ack_time)) {
49 		/* Send ACK. Note, we do not put the bucket,
50 		 * it will be released by caller.
51 		 */
52 		return TCP_TW_ACK;
53 	}
54 
55 	/* We are rate-limiting, so just release the tw sock and drop skb. */
56 	inet_twsk_put(tw);
57 	return TCP_TW_SUCCESS;
58 }
59 
60 /*
61  * * Main purpose of TIME-WAIT state is to close connection gracefully,
62  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
63  *   (and, probably, tail of data) and one or more our ACKs are lost.
64  * * What is TIME-WAIT timeout? It is associated with maximal packet
65  *   lifetime in the internet, which results in wrong conclusion, that
66  *   it is set to catch "old duplicate segments" wandering out of their path.
67  *   It is not quite correct. This timeout is calculated so that it exceeds
68  *   maximal retransmission timeout enough to allow to lose one (or more)
69  *   segments sent by peer and our ACKs. This time may be calculated from RTO.
70  * * When TIME-WAIT socket receives RST, it means that another end
71  *   finally closed and we are allowed to kill TIME-WAIT too.
72  * * Second purpose of TIME-WAIT is catching old duplicate segments.
73  *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
74  *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
75  * * If we invented some more clever way to catch duplicates
76  *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
77  *
78  * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
79  * When you compare it to RFCs, please, read section SEGMENT ARRIVES
80  * from the very beginning.
81  *
82  * NOTE. With recycling (and later with fin-wait-2) TW bucket
83  * is _not_ stateless. It means, that strictly speaking we must
84  * spinlock it. I do not want! Well, probability of misbehaviour
85  * is ridiculously low and, seems, we could use some mb() tricks
86  * to avoid misread sequence numbers, states etc.  --ANK
87  *
88  * We don't need to initialize tmp_out.sack_ok as we don't use the results
89  */
90 enum tcp_tw_status
91 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
92 			   const struct tcphdr *th)
93 {
94 	struct tcp_options_received tmp_opt;
95 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
96 	bool paws_reject = false;
97 
98 	tmp_opt.saw_tstamp = 0;
99 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
100 		tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
101 
102 		if (tmp_opt.saw_tstamp) {
103 			if (tmp_opt.rcv_tsecr)
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_timewait_check_oow_rate_limit(
120 				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
121 
122 		if (th->rst)
123 			goto kill;
124 
125 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
126 			return TCP_TW_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 			return TCP_TW_RST;
142 
143 		/* FIN arrived, enter true time-wait state. */
144 		tw->tw_substate	  = TCP_TIME_WAIT;
145 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
146 		if (tmp_opt.saw_tstamp) {
147 			tcptw->tw_ts_recent_stamp = ktime_get_seconds();
148 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
149 		}
150 
151 		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
152 		return TCP_TW_ACK;
153 	}
154 
155 	/*
156 	 *	Now real TIME-WAIT state.
157 	 *
158 	 *	RFC 1122:
159 	 *	"When a connection is [...] on TIME-WAIT state [...]
160 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
161 	 *	reopen the connection directly, if it:
162 	 *
163 	 *	(1)  assigns its initial sequence number for the new
164 	 *	connection to be larger than the largest sequence
165 	 *	number it used on the previous connection incarnation,
166 	 *	and
167 	 *
168 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
169 	 *	to be an old duplicate".
170 	 */
171 
172 	if (!paws_reject &&
173 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
174 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
175 		/* In window segment, it may be only reset or bare ack. */
176 
177 		if (th->rst) {
178 			/* This is TIME_WAIT assassination, in two flavors.
179 			 * Oh well... nobody has a sufficient solution to this
180 			 * protocol bug yet.
181 			 */
182 			if (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) {
183 kill:
184 				inet_twsk_deschedule_put(tw);
185 				return TCP_TW_SUCCESS;
186 			}
187 		} else {
188 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
189 		}
190 
191 		if (tmp_opt.saw_tstamp) {
192 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
193 			tcptw->tw_ts_recent_stamp = ktime_get_seconds();
194 		}
195 
196 		inet_twsk_put(tw);
197 		return TCP_TW_SUCCESS;
198 	}
199 
200 	/* Out of window segment.
201 
202 	   All the segments are ACKed immediately.
203 
204 	   The only exception is new SYN. We accept it, if it is
205 	   not old duplicate and we are not in danger to be killed
206 	   by delayed old duplicates. RFC check is that it has
207 	   newer sequence number works at rates <40Mbit/sec.
208 	   However, if paws works, it is reliable AND even more,
209 	   we even may relax silly seq space cutoff.
210 
211 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
212 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
213 	   we must return socket to time-wait state. It is not good,
214 	   but not fatal yet.
215 	 */
216 
217 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
218 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
219 	     (tmp_opt.saw_tstamp &&
220 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
221 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
222 		if (isn == 0)
223 			isn++;
224 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
225 		return TCP_TW_SYN;
226 	}
227 
228 	if (paws_reject)
229 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
230 
231 	if (!th->rst) {
232 		/* In this case we must reset the TIMEWAIT timer.
233 		 *
234 		 * If it is ACKless SYN it may be both old duplicate
235 		 * and new good SYN with random sequence number <rcv_nxt.
236 		 * Do not reschedule in the last case.
237 		 */
238 		if (paws_reject || th->ack)
239 			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
240 
241 		return tcp_timewait_check_oow_rate_limit(
242 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
243 	}
244 	inet_twsk_put(tw);
245 	return TCP_TW_SUCCESS;
246 }
247 EXPORT_SYMBOL(tcp_timewait_state_process);
248 
249 /*
250  * Move a socket to time-wait or dead fin-wait-2 state.
251  */
252 void tcp_time_wait(struct sock *sk, int state, int timeo)
253 {
254 	const struct inet_connection_sock *icsk = inet_csk(sk);
255 	const struct tcp_sock *tp = tcp_sk(sk);
256 	struct inet_timewait_sock *tw;
257 	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
258 
259 	tw = inet_twsk_alloc(sk, tcp_death_row, state);
260 
261 	if (tw) {
262 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
263 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
264 		struct inet_sock *inet = inet_sk(sk);
265 
266 		tw->tw_transparent	= inet->transparent;
267 		tw->tw_mark		= sk->sk_mark;
268 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
269 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
270 		tcptw->tw_snd_nxt	= tp->snd_nxt;
271 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
272 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
273 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
274 		tcptw->tw_ts_offset	= tp->tsoffset;
275 		tcptw->tw_last_oow_ack_time = 0;
276 
277 #if IS_ENABLED(CONFIG_IPV6)
278 		if (tw->tw_family == PF_INET6) {
279 			struct ipv6_pinfo *np = inet6_sk(sk);
280 
281 			tw->tw_v6_daddr = sk->sk_v6_daddr;
282 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
283 			tw->tw_tclass = np->tclass;
284 			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
285 			tw->tw_ipv6only = sk->sk_ipv6only;
286 		}
287 #endif
288 
289 #ifdef CONFIG_TCP_MD5SIG
290 		/*
291 		 * The timewait bucket does not have the key DB from the
292 		 * sock structure. We just make a quick copy of the
293 		 * md5 key being used (if indeed we are using one)
294 		 * so the timewait ack generating code has the key.
295 		 */
296 		do {
297 			tcptw->tw_md5_key = NULL;
298 			if (static_branch_unlikely(&tcp_md5_needed)) {
299 				struct tcp_md5sig_key *key;
300 
301 				key = tp->af_specific->md5_lookup(sk, sk);
302 				if (key) {
303 					tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
304 					BUG_ON(tcptw->tw_md5_key && !tcp_alloc_md5sig_pool());
305 				}
306 			}
307 		} while (0);
308 #endif
309 
310 		/* Get the TIME_WAIT timeout firing. */
311 		if (timeo < rto)
312 			timeo = rto;
313 
314 		if (state == TCP_TIME_WAIT)
315 			timeo = TCP_TIMEWAIT_LEN;
316 
317 		/* tw_timer is pinned, so we need to make sure BH are disabled
318 		 * in following section, otherwise timer handler could run before
319 		 * we complete the initialization.
320 		 */
321 		local_bh_disable();
322 		inet_twsk_schedule(tw, timeo);
323 		/* Linkage updates.
324 		 * Note that access to tw after this point is illegal.
325 		 */
326 		inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
327 		local_bh_enable();
328 	} else {
329 		/* Sorry, if we're out of memory, just CLOSE this
330 		 * socket up.  We've got bigger problems than
331 		 * non-graceful socket closings.
332 		 */
333 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
334 	}
335 
336 	tcp_update_metrics(sk);
337 	tcp_done(sk);
338 }
339 EXPORT_SYMBOL(tcp_time_wait);
340 
341 void tcp_twsk_destructor(struct sock *sk)
342 {
343 #ifdef CONFIG_TCP_MD5SIG
344 	if (static_branch_unlikely(&tcp_md5_needed)) {
345 		struct tcp_timewait_sock *twsk = tcp_twsk(sk);
346 
347 		if (twsk->tw_md5_key)
348 			kfree_rcu(twsk->tw_md5_key, rcu);
349 	}
350 #endif
351 }
352 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
353 
354 /* Warning : This function is called without sk_listener being locked.
355  * Be sure to read socket fields once, as their value could change under us.
356  */
357 void tcp_openreq_init_rwin(struct request_sock *req,
358 			   const struct sock *sk_listener,
359 			   const struct dst_entry *dst)
360 {
361 	struct inet_request_sock *ireq = inet_rsk(req);
362 	const struct tcp_sock *tp = tcp_sk(sk_listener);
363 	int full_space = tcp_full_space(sk_listener);
364 	u32 window_clamp;
365 	__u8 rcv_wscale;
366 	u32 rcv_wnd;
367 	int mss;
368 
369 	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
370 	window_clamp = READ_ONCE(tp->window_clamp);
371 	/* Set this up on the first call only */
372 	req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
373 
374 	/* limit the window selection if the user enforce a smaller rx buffer */
375 	if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
376 	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
377 		req->rsk_window_clamp = full_space;
378 
379 	rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
380 	if (rcv_wnd == 0)
381 		rcv_wnd = dst_metric(dst, RTAX_INITRWND);
382 	else if (full_space < rcv_wnd * mss)
383 		full_space = rcv_wnd * mss;
384 
385 	/* tcp_full_space because it is guaranteed to be the first packet */
386 	tcp_select_initial_window(sk_listener, full_space,
387 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
388 		&req->rsk_rcv_wnd,
389 		&req->rsk_window_clamp,
390 		ireq->wscale_ok,
391 		&rcv_wscale,
392 		rcv_wnd);
393 	ireq->rcv_wscale = rcv_wscale;
394 }
395 EXPORT_SYMBOL(tcp_openreq_init_rwin);
396 
397 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
398 				  const struct request_sock *req)
399 {
400 	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
401 }
402 
403 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
404 {
405 	struct inet_connection_sock *icsk = inet_csk(sk);
406 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
407 	bool ca_got_dst = false;
408 
409 	if (ca_key != TCP_CA_UNSPEC) {
410 		const struct tcp_congestion_ops *ca;
411 
412 		rcu_read_lock();
413 		ca = tcp_ca_find_key(ca_key);
414 		if (likely(ca && try_module_get(ca->owner))) {
415 			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
416 			icsk->icsk_ca_ops = ca;
417 			ca_got_dst = true;
418 		}
419 		rcu_read_unlock();
420 	}
421 
422 	/* If no valid choice made yet, assign current system default ca. */
423 	if (!ca_got_dst &&
424 	    (!icsk->icsk_ca_setsockopt ||
425 	     !try_module_get(icsk->icsk_ca_ops->owner)))
426 		tcp_assign_congestion_control(sk);
427 
428 	tcp_set_ca_state(sk, TCP_CA_Open);
429 }
430 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
431 
432 static void smc_check_reset_syn_req(struct tcp_sock *oldtp,
433 				    struct request_sock *req,
434 				    struct tcp_sock *newtp)
435 {
436 #if IS_ENABLED(CONFIG_SMC)
437 	struct inet_request_sock *ireq;
438 
439 	if (static_branch_unlikely(&tcp_have_smc)) {
440 		ireq = inet_rsk(req);
441 		if (oldtp->syn_smc && !ireq->smc_ok)
442 			newtp->syn_smc = 0;
443 	}
444 #endif
445 }
446 
447 /* This is not only more efficient than what we used to do, it eliminates
448  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
449  *
450  * Actually, we could lots of memory writes here. tp of listening
451  * socket contains all necessary default parameters.
452  */
453 struct sock *tcp_create_openreq_child(const struct sock *sk,
454 				      struct request_sock *req,
455 				      struct sk_buff *skb)
456 {
457 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
458 	const struct inet_request_sock *ireq = inet_rsk(req);
459 	struct tcp_request_sock *treq = tcp_rsk(req);
460 	struct inet_connection_sock *newicsk;
461 	struct tcp_sock *oldtp, *newtp;
462 
463 	if (!newsk)
464 		return NULL;
465 
466 	newicsk = inet_csk(newsk);
467 	newtp = tcp_sk(newsk);
468 	oldtp = tcp_sk(sk);
469 
470 	smc_check_reset_syn_req(oldtp, req, newtp);
471 
472 	/* Now setup tcp_sock */
473 	newtp->pred_flags = 0;
474 
475 	newtp->rcv_wup = newtp->copied_seq =
476 	newtp->rcv_nxt = treq->rcv_isn + 1;
477 	newtp->segs_in = 1;
478 
479 	newtp->snd_sml = newtp->snd_una =
480 	newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
481 
482 	INIT_LIST_HEAD(&newtp->tsq_node);
483 	INIT_LIST_HEAD(&newtp->tsorted_sent_queue);
484 
485 	tcp_init_wl(newtp, treq->rcv_isn);
486 
487 	minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
488 	newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
489 
490 	newtp->lsndtime = tcp_jiffies32;
491 	newsk->sk_txhash = treq->txhash;
492 	newtp->total_retrans = req->num_retrans;
493 
494 	tcp_init_xmit_timers(newsk);
495 	newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
496 
497 	if (sock_flag(newsk, SOCK_KEEPOPEN))
498 		inet_csk_reset_keepalive_timer(newsk,
499 					       keepalive_time_when(newtp));
500 
501 	newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
502 	newtp->rx_opt.sack_ok = ireq->sack_ok;
503 	newtp->window_clamp = req->rsk_window_clamp;
504 	newtp->rcv_ssthresh = req->rsk_rcv_wnd;
505 	newtp->rcv_wnd = req->rsk_rcv_wnd;
506 	newtp->rx_opt.wscale_ok = ireq->wscale_ok;
507 	if (newtp->rx_opt.wscale_ok) {
508 		newtp->rx_opt.snd_wscale = ireq->snd_wscale;
509 		newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
510 	} else {
511 		newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
512 		newtp->window_clamp = min(newtp->window_clamp, 65535U);
513 	}
514 	newtp->snd_wnd = ntohs(tcp_hdr(skb)->window) << newtp->rx_opt.snd_wscale;
515 	newtp->max_window = newtp->snd_wnd;
516 
517 	if (newtp->rx_opt.tstamp_ok) {
518 		newtp->rx_opt.ts_recent = req->ts_recent;
519 		newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
520 		newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
521 	} else {
522 		newtp->rx_opt.ts_recent_stamp = 0;
523 		newtp->tcp_header_len = sizeof(struct tcphdr);
524 	}
525 	newtp->tsoffset = treq->ts_off;
526 #ifdef CONFIG_TCP_MD5SIG
527 	newtp->md5sig_info = NULL;	/*XXX*/
528 	if (newtp->af_specific->md5_lookup(sk, newsk))
529 		newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
530 #endif
531 	if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
532 		newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
533 	newtp->rx_opt.mss_clamp = req->mss;
534 	tcp_ecn_openreq_child(newtp, req);
535 	newtp->fastopen_req = NULL;
536 	newtp->fastopen_rsk = NULL;
537 
538 	__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
539 
540 	return newsk;
541 }
542 EXPORT_SYMBOL(tcp_create_openreq_child);
543 
544 /*
545  * Process an incoming packet for SYN_RECV sockets represented as a
546  * request_sock. Normally sk is the listener socket but for TFO it
547  * points to the child socket.
548  *
549  * XXX (TFO) - The current impl contains a special check for ack
550  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
551  *
552  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
553  */
554 
555 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
556 			   struct request_sock *req,
557 			   bool fastopen, bool *req_stolen)
558 {
559 	struct tcp_options_received tmp_opt;
560 	struct sock *child;
561 	const struct tcphdr *th = tcp_hdr(skb);
562 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
563 	bool paws_reject = false;
564 	bool own_req;
565 
566 	tmp_opt.saw_tstamp = 0;
567 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
568 		tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
569 
570 		if (tmp_opt.saw_tstamp) {
571 			tmp_opt.ts_recent = req->ts_recent;
572 			if (tmp_opt.rcv_tsecr)
573 				tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
574 			/* We do not store true stamp, but it is not required,
575 			 * it can be estimated (approximately)
576 			 * from another data.
577 			 */
578 			tmp_opt.ts_recent_stamp = ktime_get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
579 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
580 		}
581 	}
582 
583 	/* Check for pure retransmitted SYN. */
584 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
585 	    flg == TCP_FLAG_SYN &&
586 	    !paws_reject) {
587 		/*
588 		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
589 		 * this case on figure 6 and figure 8, but formal
590 		 * protocol description says NOTHING.
591 		 * To be more exact, it says that we should send ACK,
592 		 * because this segment (at least, if it has no data)
593 		 * is out of window.
594 		 *
595 		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
596 		 *  describe SYN-RECV state. All the description
597 		 *  is wrong, we cannot believe to it and should
598 		 *  rely only on common sense and implementation
599 		 *  experience.
600 		 *
601 		 * Enforce "SYN-ACK" according to figure 8, figure 6
602 		 * of RFC793, fixed by RFC1122.
603 		 *
604 		 * Note that even if there is new data in the SYN packet
605 		 * they will be thrown away too.
606 		 *
607 		 * Reset timer after retransmitting SYNACK, similar to
608 		 * the idea of fast retransmit in recovery.
609 		 */
610 		if (!tcp_oow_rate_limited(sock_net(sk), skb,
611 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
612 					  &tcp_rsk(req)->last_oow_ack_time) &&
613 
614 		    !inet_rtx_syn_ack(sk, req)) {
615 			unsigned long expires = jiffies;
616 
617 			expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
618 				       TCP_RTO_MAX);
619 			if (!fastopen)
620 				mod_timer_pending(&req->rsk_timer, expires);
621 			else
622 				req->rsk_timer.expires = expires;
623 		}
624 		return NULL;
625 	}
626 
627 	/* Further reproduces section "SEGMENT ARRIVES"
628 	   for state SYN-RECEIVED of RFC793.
629 	   It is broken, however, it does not work only
630 	   when SYNs are crossed.
631 
632 	   You would think that SYN crossing is impossible here, since
633 	   we should have a SYN_SENT socket (from connect()) on our end,
634 	   but this is not true if the crossed SYNs were sent to both
635 	   ends by a malicious third party.  We must defend against this,
636 	   and to do that we first verify the ACK (as per RFC793, page
637 	   36) and reset if it is invalid.  Is this a true full defense?
638 	   To convince ourselves, let us consider a way in which the ACK
639 	   test can still pass in this 'malicious crossed SYNs' case.
640 	   Malicious sender sends identical SYNs (and thus identical sequence
641 	   numbers) to both A and B:
642 
643 		A: gets SYN, seq=7
644 		B: gets SYN, seq=7
645 
646 	   By our good fortune, both A and B select the same initial
647 	   send sequence number of seven :-)
648 
649 		A: sends SYN|ACK, seq=7, ack_seq=8
650 		B: sends SYN|ACK, seq=7, ack_seq=8
651 
652 	   So we are now A eating this SYN|ACK, ACK test passes.  So
653 	   does sequence test, SYN is truncated, and thus we consider
654 	   it a bare ACK.
655 
656 	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
657 	   bare ACK.  Otherwise, we create an established connection.  Both
658 	   ends (listening sockets) accept the new incoming connection and try
659 	   to talk to each other. 8-)
660 
661 	   Note: This case is both harmless, and rare.  Possibility is about the
662 	   same as us discovering intelligent life on another plant tomorrow.
663 
664 	   But generally, we should (RFC lies!) to accept ACK
665 	   from SYNACK both here and in tcp_rcv_state_process().
666 	   tcp_rcv_state_process() does not, hence, we do not too.
667 
668 	   Note that the case is absolutely generic:
669 	   we cannot optimize anything here without
670 	   violating protocol. All the checks must be made
671 	   before attempt to create socket.
672 	 */
673 
674 	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
675 	 *                  and the incoming segment acknowledges something not yet
676 	 *                  sent (the segment carries an unacceptable ACK) ...
677 	 *                  a reset is sent."
678 	 *
679 	 * Invalid ACK: reset will be sent by listening socket.
680 	 * Note that the ACK validity check for a Fast Open socket is done
681 	 * elsewhere and is checked directly against the child socket rather
682 	 * than req because user data may have been sent out.
683 	 */
684 	if ((flg & TCP_FLAG_ACK) && !fastopen &&
685 	    (TCP_SKB_CB(skb)->ack_seq !=
686 	     tcp_rsk(req)->snt_isn + 1))
687 		return sk;
688 
689 	/* Also, it would be not so bad idea to check rcv_tsecr, which
690 	 * is essentially ACK extension and too early or too late values
691 	 * should cause reset in unsynchronized states.
692 	 */
693 
694 	/* RFC793: "first check sequence number". */
695 
696 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
697 					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
698 		/* Out of window: send ACK and drop. */
699 		if (!(flg & TCP_FLAG_RST) &&
700 		    !tcp_oow_rate_limited(sock_net(sk), skb,
701 					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
702 					  &tcp_rsk(req)->last_oow_ack_time))
703 			req->rsk_ops->send_ack(sk, skb, req);
704 		if (paws_reject)
705 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
706 		return NULL;
707 	}
708 
709 	/* In sequence, PAWS is OK. */
710 
711 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
712 		req->ts_recent = tmp_opt.rcv_tsval;
713 
714 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
715 		/* Truncate SYN, it is out of window starting
716 		   at tcp_rsk(req)->rcv_isn + 1. */
717 		flg &= ~TCP_FLAG_SYN;
718 	}
719 
720 	/* RFC793: "second check the RST bit" and
721 	 *	   "fourth, check the SYN bit"
722 	 */
723 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
724 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
725 		goto embryonic_reset;
726 	}
727 
728 	/* ACK sequence verified above, just make sure ACK is
729 	 * set.  If ACK not set, just silently drop the packet.
730 	 *
731 	 * XXX (TFO) - if we ever allow "data after SYN", the
732 	 * following check needs to be removed.
733 	 */
734 	if (!(flg & TCP_FLAG_ACK))
735 		return NULL;
736 
737 	/* For Fast Open no more processing is needed (sk is the
738 	 * child socket).
739 	 */
740 	if (fastopen)
741 		return sk;
742 
743 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
744 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
745 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
746 		inet_rsk(req)->acked = 1;
747 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
748 		return NULL;
749 	}
750 
751 	/* OK, ACK is valid, create big socket and
752 	 * feed this segment to it. It will repeat all
753 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
754 	 * ESTABLISHED STATE. If it will be dropped after
755 	 * socket is created, wait for troubles.
756 	 */
757 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
758 							 req, &own_req);
759 	if (!child)
760 		goto listen_overflow;
761 
762 	sock_rps_save_rxhash(child, skb);
763 	tcp_synack_rtt_meas(child, req);
764 	*req_stolen = !own_req;
765 	return inet_csk_complete_hashdance(sk, child, req, own_req);
766 
767 listen_overflow:
768 	if (!sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow) {
769 		inet_rsk(req)->acked = 1;
770 		return NULL;
771 	}
772 
773 embryonic_reset:
774 	if (!(flg & TCP_FLAG_RST)) {
775 		/* Received a bad SYN pkt - for TFO We try not to reset
776 		 * the local connection unless it's really necessary to
777 		 * avoid becoming vulnerable to outside attack aiming at
778 		 * resetting legit local connections.
779 		 */
780 		req->rsk_ops->send_reset(sk, skb);
781 	} else if (fastopen) { /* received a valid RST pkt */
782 		reqsk_fastopen_remove(sk, req, true);
783 		tcp_reset(sk);
784 	}
785 	if (!fastopen) {
786 		inet_csk_reqsk_queue_drop(sk, req);
787 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
788 	}
789 	return NULL;
790 }
791 EXPORT_SYMBOL(tcp_check_req);
792 
793 /*
794  * Queue segment on the new socket if the new socket is active,
795  * otherwise we just shortcircuit this and continue with
796  * the new socket.
797  *
798  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
799  * when entering. But other states are possible due to a race condition
800  * where after __inet_lookup_established() fails but before the listener
801  * locked is obtained, other packets cause the same connection to
802  * be created.
803  */
804 
805 int tcp_child_process(struct sock *parent, struct sock *child,
806 		      struct sk_buff *skb)
807 {
808 	int ret = 0;
809 	int state = child->sk_state;
810 
811 	/* record NAPI ID of child */
812 	sk_mark_napi_id(child, skb);
813 
814 	tcp_segs_in(tcp_sk(child), skb);
815 	if (!sock_owned_by_user(child)) {
816 		ret = tcp_rcv_state_process(child, skb);
817 		/* Wakeup parent, send SIGIO */
818 		if (state == TCP_SYN_RECV && child->sk_state != state)
819 			parent->sk_data_ready(parent);
820 	} else {
821 		/* Alas, it is possible again, because we do lookup
822 		 * in main socket hash table and lock on listening
823 		 * socket does not protect us more.
824 		 */
825 		__sk_add_backlog(child, skb);
826 	}
827 
828 	bh_unlock_sock(child);
829 	sock_put(child);
830 	return ret;
831 }
832 EXPORT_SYMBOL(tcp_child_process);
833