xref: /openbmc/linux/net/ipv4/tcp_minisocks.c (revision 976fa9a3)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Implementation of the Transmission Control Protocol(TCP).
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
11  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
12  *		Florian La Roche, <flla@stud.uni-sb.de>
13  *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14  *		Linus Torvalds, <torvalds@cs.helsinki.fi>
15  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
16  *		Matthew Dillon, <dillon@apollo.west.oic.com>
17  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18  *		Jorge Cwik, <jorge@laser.satlink.net>
19  */
20 
21 #include <linux/mm.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <linux/sysctl.h>
25 #include <linux/workqueue.h>
26 #include <net/tcp.h>
27 #include <net/inet_common.h>
28 #include <net/xfrm.h>
29 
30 int sysctl_tcp_syncookies __read_mostly = 1;
31 EXPORT_SYMBOL(sysctl_tcp_syncookies);
32 
33 int sysctl_tcp_abort_on_overflow __read_mostly;
34 
35 struct inet_timewait_death_row tcp_death_row = {
36 	.sysctl_max_tw_buckets = NR_FILE * 2,
37 	.hashinfo	= &tcp_hashinfo,
38 };
39 EXPORT_SYMBOL_GPL(tcp_death_row);
40 
41 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
42 {
43 	if (seq == s_win)
44 		return true;
45 	if (after(end_seq, s_win) && before(seq, e_win))
46 		return true;
47 	return seq == e_win && seq == end_seq;
48 }
49 
50 static enum tcp_tw_status
51 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
52 				  const struct sk_buff *skb, int mib_idx)
53 {
54 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
55 
56 	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
57 				  &tcptw->tw_last_oow_ack_time)) {
58 		/* Send ACK. Note, we do not put the bucket,
59 		 * it will be released by caller.
60 		 */
61 		return TCP_TW_ACK;
62 	}
63 
64 	/* We are rate-limiting, so just release the tw sock and drop skb. */
65 	inet_twsk_put(tw);
66 	return TCP_TW_SUCCESS;
67 }
68 
69 /*
70  * * Main purpose of TIME-WAIT state is to close connection gracefully,
71  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
72  *   (and, probably, tail of data) and one or more our ACKs are lost.
73  * * What is TIME-WAIT timeout? It is associated with maximal packet
74  *   lifetime in the internet, which results in wrong conclusion, that
75  *   it is set to catch "old duplicate segments" wandering out of their path.
76  *   It is not quite correct. This timeout is calculated so that it exceeds
77  *   maximal retransmission timeout enough to allow to lose one (or more)
78  *   segments sent by peer and our ACKs. This time may be calculated from RTO.
79  * * When TIME-WAIT socket receives RST, it means that another end
80  *   finally closed and we are allowed to kill TIME-WAIT too.
81  * * Second purpose of TIME-WAIT is catching old duplicate segments.
82  *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
83  *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
84  * * If we invented some more clever way to catch duplicates
85  *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
86  *
87  * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
88  * When you compare it to RFCs, please, read section SEGMENT ARRIVES
89  * from the very beginning.
90  *
91  * NOTE. With recycling (and later with fin-wait-2) TW bucket
92  * is _not_ stateless. It means, that strictly speaking we must
93  * spinlock it. I do not want! Well, probability of misbehaviour
94  * is ridiculously low and, seems, we could use some mb() tricks
95  * to avoid misread sequence numbers, states etc.  --ANK
96  *
97  * We don't need to initialize tmp_out.sack_ok as we don't use the results
98  */
99 enum tcp_tw_status
100 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
101 			   const struct tcphdr *th)
102 {
103 	struct tcp_options_received tmp_opt;
104 	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
105 	bool paws_reject = false;
106 
107 	tmp_opt.saw_tstamp = 0;
108 	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
109 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
110 
111 		if (tmp_opt.saw_tstamp) {
112 			tmp_opt.rcv_tsecr	-= tcptw->tw_ts_offset;
113 			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
114 			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
115 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
116 		}
117 	}
118 
119 	if (tw->tw_substate == TCP_FIN_WAIT2) {
120 		/* Just repeat all the checks of tcp_rcv_state_process() */
121 
122 		/* Out of window, send ACK */
123 		if (paws_reject ||
124 		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
125 				   tcptw->tw_rcv_nxt,
126 				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
127 			return tcp_timewait_check_oow_rate_limit(
128 				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
129 
130 		if (th->rst)
131 			goto kill;
132 
133 		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
134 			goto kill_with_rst;
135 
136 		/* Dup ACK? */
137 		if (!th->ack ||
138 		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
139 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
140 			inet_twsk_put(tw);
141 			return TCP_TW_SUCCESS;
142 		}
143 
144 		/* New data or FIN. If new data arrive after half-duplex close,
145 		 * reset.
146 		 */
147 		if (!th->fin ||
148 		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
149 kill_with_rst:
150 			inet_twsk_deschedule(tw);
151 			inet_twsk_put(tw);
152 			return TCP_TW_RST;
153 		}
154 
155 		/* FIN arrived, enter true time-wait state. */
156 		tw->tw_substate	  = TCP_TIME_WAIT;
157 		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
158 		if (tmp_opt.saw_tstamp) {
159 			tcptw->tw_ts_recent_stamp = get_seconds();
160 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
161 		}
162 
163 		if (tcp_death_row.sysctl_tw_recycle &&
164 		    tcptw->tw_ts_recent_stamp &&
165 		    tcp_tw_remember_stamp(tw))
166 			inet_twsk_schedule(tw, tw->tw_timeout);
167 		else
168 			inet_twsk_schedule(tw, TCP_TIMEWAIT_LEN);
169 		return TCP_TW_ACK;
170 	}
171 
172 	/*
173 	 *	Now real TIME-WAIT state.
174 	 *
175 	 *	RFC 1122:
176 	 *	"When a connection is [...] on TIME-WAIT state [...]
177 	 *	[a TCP] MAY accept a new SYN from the remote TCP to
178 	 *	reopen the connection directly, if it:
179 	 *
180 	 *	(1)  assigns its initial sequence number for the new
181 	 *	connection to be larger than the largest sequence
182 	 *	number it used on the previous connection incarnation,
183 	 *	and
184 	 *
185 	 *	(2)  returns to TIME-WAIT state if the SYN turns out
186 	 *	to be an old duplicate".
187 	 */
188 
189 	if (!paws_reject &&
190 	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
191 	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
192 		/* In window segment, it may be only reset or bare ack. */
193 
194 		if (th->rst) {
195 			/* This is TIME_WAIT assassination, in two flavors.
196 			 * Oh well... nobody has a sufficient solution to this
197 			 * protocol bug yet.
198 			 */
199 			if (sysctl_tcp_rfc1337 == 0) {
200 kill:
201 				inet_twsk_deschedule(tw);
202 				inet_twsk_put(tw);
203 				return TCP_TW_SUCCESS;
204 			}
205 		}
206 		inet_twsk_schedule(tw, TCP_TIMEWAIT_LEN);
207 
208 		if (tmp_opt.saw_tstamp) {
209 			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
210 			tcptw->tw_ts_recent_stamp = get_seconds();
211 		}
212 
213 		inet_twsk_put(tw);
214 		return TCP_TW_SUCCESS;
215 	}
216 
217 	/* Out of window segment.
218 
219 	   All the segments are ACKed immediately.
220 
221 	   The only exception is new SYN. We accept it, if it is
222 	   not old duplicate and we are not in danger to be killed
223 	   by delayed old duplicates. RFC check is that it has
224 	   newer sequence number works at rates <40Mbit/sec.
225 	   However, if paws works, it is reliable AND even more,
226 	   we even may relax silly seq space cutoff.
227 
228 	   RED-PEN: we violate main RFC requirement, if this SYN will appear
229 	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
230 	   we must return socket to time-wait state. It is not good,
231 	   but not fatal yet.
232 	 */
233 
234 	if (th->syn && !th->rst && !th->ack && !paws_reject &&
235 	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
236 	     (tmp_opt.saw_tstamp &&
237 	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
238 		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
239 		if (isn == 0)
240 			isn++;
241 		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
242 		return TCP_TW_SYN;
243 	}
244 
245 	if (paws_reject)
246 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
247 
248 	if (!th->rst) {
249 		/* In this case we must reset the TIMEWAIT timer.
250 		 *
251 		 * If it is ACKless SYN it may be both old duplicate
252 		 * and new good SYN with random sequence number <rcv_nxt.
253 		 * Do not reschedule in the last case.
254 		 */
255 		if (paws_reject || th->ack)
256 			inet_twsk_schedule(tw, TCP_TIMEWAIT_LEN);
257 
258 		return tcp_timewait_check_oow_rate_limit(
259 			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
260 	}
261 	inet_twsk_put(tw);
262 	return TCP_TW_SUCCESS;
263 }
264 EXPORT_SYMBOL(tcp_timewait_state_process);
265 
266 /*
267  * Move a socket to time-wait or dead fin-wait-2 state.
268  */
269 void tcp_time_wait(struct sock *sk, int state, int timeo)
270 {
271 	const struct inet_connection_sock *icsk = inet_csk(sk);
272 	const struct tcp_sock *tp = tcp_sk(sk);
273 	struct inet_timewait_sock *tw;
274 	bool recycle_ok = false;
275 
276 	if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
277 		recycle_ok = tcp_remember_stamp(sk);
278 
279 	tw = inet_twsk_alloc(sk, &tcp_death_row, state);
280 
281 	if (tw) {
282 		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
283 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
284 		struct inet_sock *inet = inet_sk(sk);
285 
286 		tw->tw_transparent	= inet->transparent;
287 		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
288 		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
289 		tcptw->tw_snd_nxt	= tp->snd_nxt;
290 		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
291 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
292 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
293 		tcptw->tw_ts_offset	= tp->tsoffset;
294 		tcptw->tw_last_oow_ack_time = 0;
295 
296 #if IS_ENABLED(CONFIG_IPV6)
297 		if (tw->tw_family == PF_INET6) {
298 			struct ipv6_pinfo *np = inet6_sk(sk);
299 
300 			tw->tw_v6_daddr = sk->sk_v6_daddr;
301 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
302 			tw->tw_tclass = np->tclass;
303 			tw->tw_flowlabel = np->flow_label >> 12;
304 			tw->tw_ipv6only = sk->sk_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) {
320 				tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
321 				if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
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, timeo);
343 		inet_twsk_put(tw);
344 	} else {
345 		/* Sorry, if we're out of memory, just CLOSE this
346 		 * socket up.  We've got bigger problems than
347 		 * non-graceful socket closings.
348 		 */
349 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
350 	}
351 
352 	tcp_update_metrics(sk);
353 	tcp_done(sk);
354 }
355 
356 void tcp_twsk_destructor(struct sock *sk)
357 {
358 #ifdef CONFIG_TCP_MD5SIG
359 	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
360 
361 	if (twsk->tw_md5_key)
362 		kfree_rcu(twsk->tw_md5_key, rcu);
363 #endif
364 }
365 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
366 
367 void tcp_openreq_init_rwin(struct request_sock *req,
368 			   struct sock *sk, struct dst_entry *dst)
369 {
370 	struct inet_request_sock *ireq = inet_rsk(req);
371 	struct tcp_sock *tp = tcp_sk(sk);
372 	__u8 rcv_wscale;
373 	int mss = dst_metric_advmss(dst);
374 
375 	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
376 		mss = tp->rx_opt.user_mss;
377 
378 	/* Set this up on the first call only */
379 	req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
380 
381 	/* limit the window selection if the user enforce a smaller rx buffer */
382 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
383 	    (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0))
384 		req->window_clamp = tcp_full_space(sk);
385 
386 	/* tcp_full_space because it is guaranteed to be the first packet */
387 	tcp_select_initial_window(tcp_full_space(sk),
388 		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
389 		&req->rcv_wnd,
390 		&req->window_clamp,
391 		ireq->wscale_ok,
392 		&rcv_wscale,
393 		dst_metric(dst, RTAX_INITRWND));
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 (!ca_got_dst && !try_module_get(icsk->icsk_ca_ops->owner))
424 		tcp_assign_congestion_control(sk);
425 
426 	tcp_set_ca_state(sk, TCP_CA_Open);
427 }
428 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
429 
430 /* This is not only more efficient than what we used to do, it eliminates
431  * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
432  *
433  * Actually, we could lots of memory writes here. tp of listening
434  * socket contains all necessary default parameters.
435  */
436 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
437 {
438 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
439 
440 	if (newsk) {
441 		const struct inet_request_sock *ireq = inet_rsk(req);
442 		struct tcp_request_sock *treq = tcp_rsk(req);
443 		struct inet_connection_sock *newicsk = inet_csk(newsk);
444 		struct tcp_sock *newtp = tcp_sk(newsk);
445 
446 		/* Now setup tcp_sock */
447 		newtp->pred_flags = 0;
448 
449 		newtp->rcv_wup = newtp->copied_seq =
450 		newtp->rcv_nxt = treq->rcv_isn + 1;
451 
452 		newtp->snd_sml = newtp->snd_una =
453 		newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
454 
455 		tcp_prequeue_init(newtp);
456 		INIT_LIST_HEAD(&newtp->tsq_node);
457 
458 		tcp_init_wl(newtp, treq->rcv_isn);
459 
460 		newtp->srtt_us = 0;
461 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
462 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
463 
464 		newtp->packets_out = 0;
465 		newtp->retrans_out = 0;
466 		newtp->sacked_out = 0;
467 		newtp->fackets_out = 0;
468 		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
469 		tcp_enable_early_retrans(newtp);
470 		newtp->tlp_high_seq = 0;
471 		newtp->lsndtime = treq->snt_synack;
472 		newtp->last_oow_ack_time = 0;
473 		newtp->total_retrans = req->num_retrans;
474 
475 		/* So many TCP implementations out there (incorrectly) count the
476 		 * initial SYN frame in their delayed-ACK and congestion control
477 		 * algorithms that we must have the following bandaid to talk
478 		 * efficiently to them.  -DaveM
479 		 */
480 		newtp->snd_cwnd = TCP_INIT_CWND;
481 		newtp->snd_cwnd_cnt = 0;
482 
483 		tcp_init_xmit_timers(newsk);
484 		__skb_queue_head_init(&newtp->out_of_order_queue);
485 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
486 
487 		newtp->rx_opt.saw_tstamp = 0;
488 
489 		newtp->rx_opt.dsack = 0;
490 		newtp->rx_opt.num_sacks = 0;
491 
492 		newtp->urg_data = 0;
493 
494 		if (sock_flag(newsk, SOCK_KEEPOPEN))
495 			inet_csk_reset_keepalive_timer(newsk,
496 						       keepalive_time_when(newtp));
497 
498 		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
499 		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
500 			if (sysctl_tcp_fack)
501 				tcp_enable_fack(newtp);
502 		}
503 		newtp->window_clamp = req->window_clamp;
504 		newtp->rcv_ssthresh = req->rcv_wnd;
505 		newtp->rcv_wnd = req->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) <<
515 				  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 = 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 		newtp->tsoffset = 0;
527 #ifdef CONFIG_TCP_MD5SIG
528 		newtp->md5sig_info = NULL;	/*XXX*/
529 		if (newtp->af_specific->md5_lookup(sk, newsk))
530 			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
531 #endif
532 		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
533 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
534 		newtp->rx_opt.mss_clamp = req->mss;
535 		tcp_ecn_openreq_child(newtp, req);
536 		newtp->fastopen_rsk = NULL;
537 		newtp->syn_data_acked = 0;
538 
539 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
540 	}
541 	return newsk;
542 }
543 EXPORT_SYMBOL(tcp_create_openreq_child);
544 
545 /*
546  * Process an incoming packet for SYN_RECV sockets represented as a
547  * request_sock. Normally sk is the listener socket but for TFO it
548  * points to the child socket.
549  *
550  * XXX (TFO) - The current impl contains a special check for ack
551  * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
552  *
553  * We don't need to initialize tmp_opt.sack_ok as we don't use the results
554  */
555 
556 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
557 			   struct request_sock *req,
558 			   bool fastopen)
559 {
560 	struct tcp_options_received tmp_opt;
561 	struct sock *child;
562 	const struct tcphdr *th = tcp_hdr(skb);
563 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
564 	bool paws_reject = false;
565 
566 	BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
567 
568 	tmp_opt.saw_tstamp = 0;
569 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
570 		tcp_parse_options(skb, &tmp_opt, 0, NULL);
571 
572 		if (tmp_opt.saw_tstamp) {
573 			tmp_opt.ts_recent = req->ts_recent;
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 = 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->rcv_wnd)) {
698 		/* Out of window: send ACK and drop. */
699 		if (!(flg & TCP_FLAG_RST))
700 			req->rsk_ops->send_ack(sk, skb, req);
701 		if (paws_reject)
702 			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
703 		return NULL;
704 	}
705 
706 	/* In sequence, PAWS is OK. */
707 
708 	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
709 		req->ts_recent = tmp_opt.rcv_tsval;
710 
711 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
712 		/* Truncate SYN, it is out of window starting
713 		   at tcp_rsk(req)->rcv_isn + 1. */
714 		flg &= ~TCP_FLAG_SYN;
715 	}
716 
717 	/* RFC793: "second check the RST bit" and
718 	 *	   "fourth, check the SYN bit"
719 	 */
720 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
721 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
722 		goto embryonic_reset;
723 	}
724 
725 	/* ACK sequence verified above, just make sure ACK is
726 	 * set.  If ACK not set, just silently drop the packet.
727 	 *
728 	 * XXX (TFO) - if we ever allow "data after SYN", the
729 	 * following check needs to be removed.
730 	 */
731 	if (!(flg & TCP_FLAG_ACK))
732 		return NULL;
733 
734 	/* For Fast Open no more processing is needed (sk is the
735 	 * child socket).
736 	 */
737 	if (fastopen)
738 		return sk;
739 
740 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
741 	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
742 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
743 		inet_rsk(req)->acked = 1;
744 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
745 		return NULL;
746 	}
747 
748 	/* OK, ACK is valid, create big socket and
749 	 * feed this segment to it. It will repeat all
750 	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
751 	 * ESTABLISHED STATE. If it will be dropped after
752 	 * socket is created, wait for troubles.
753 	 */
754 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
755 	if (!child)
756 		goto listen_overflow;
757 
758 	inet_csk_reqsk_queue_drop(sk, req);
759 	inet_csk_reqsk_queue_add(sk, req, child);
760 	/* Warning: caller must not call reqsk_put(req);
761 	 * child stole last reference on it.
762 	 */
763 	return child;
764 
765 listen_overflow:
766 	if (!sysctl_tcp_abort_on_overflow) {
767 		inet_rsk(req)->acked = 1;
768 		return NULL;
769 	}
770 
771 embryonic_reset:
772 	if (!(flg & TCP_FLAG_RST)) {
773 		/* Received a bad SYN pkt - for TFO We try not to reset
774 		 * the local connection unless it's really necessary to
775 		 * avoid becoming vulnerable to outside attack aiming at
776 		 * resetting legit local connections.
777 		 */
778 		req->rsk_ops->send_reset(sk, skb);
779 	} else if (fastopen) { /* received a valid RST pkt */
780 		reqsk_fastopen_remove(sk, req, true);
781 		tcp_reset(sk);
782 	}
783 	if (!fastopen) {
784 		inet_csk_reqsk_queue_drop(sk, req);
785 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
786 	}
787 	return NULL;
788 }
789 EXPORT_SYMBOL(tcp_check_req);
790 
791 /*
792  * Queue segment on the new socket if the new socket is active,
793  * otherwise we just shortcircuit this and continue with
794  * the new socket.
795  *
796  * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
797  * when entering. But other states are possible due to a race condition
798  * where after __inet_lookup_established() fails but before the listener
799  * locked is obtained, other packets cause the same connection to
800  * be created.
801  */
802 
803 int tcp_child_process(struct sock *parent, struct sock *child,
804 		      struct sk_buff *skb)
805 {
806 	int ret = 0;
807 	int state = child->sk_state;
808 
809 	if (!sock_owned_by_user(child)) {
810 		ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
811 					    skb->len);
812 		/* Wakeup parent, send SIGIO */
813 		if (state == TCP_SYN_RECV && child->sk_state != state)
814 			parent->sk_data_ready(parent);
815 	} else {
816 		/* Alas, it is possible again, because we do lookup
817 		 * in main socket hash table and lock on listening
818 		 * socket does not protect us more.
819 		 */
820 		__sk_add_backlog(child, skb);
821 	}
822 
823 	bh_unlock_sock(child);
824 	sock_put(child);
825 	return ret;
826 }
827 EXPORT_SYMBOL(tcp_child_process);
828