Linux Audio

Check our new training course

Loading...
v6.9.4
  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 <net/tcp.h>
 
 23#include <net/xfrm.h>
 24#include <net/busy_poll.h>
 25
 26static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
 27{
 28	if (seq == s_win)
 29		return true;
 30	if (after(end_seq, s_win) && before(seq, e_win))
 31		return true;
 32	return seq == e_win && seq == end_seq;
 33}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 34
 35static enum tcp_tw_status
 36tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
 37				  const struct sk_buff *skb, int mib_idx)
 38{
 39	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
 
 
 
 40
 41	if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
 42				  &tcptw->tw_last_oow_ack_time)) {
 43		/* Send ACK. Note, we do not put the bucket,
 44		 * it will be released by caller.
 45		 */
 46		return TCP_TW_ACK;
 
 
 
 
 
 47	}
 48
 49	/* We are rate-limiting, so just release the tw sock and drop skb. */
 50	inet_twsk_put(tw);
 51	return TCP_TW_SUCCESS;
 52}
 53
 54static void twsk_rcv_nxt_update(struct tcp_timewait_sock *tcptw, u32 seq)
 55{
 56#ifdef CONFIG_TCP_AO
 57	struct tcp_ao_info *ao;
 58
 59	ao = rcu_dereference(tcptw->ao_info);
 60	if (unlikely(ao && seq < tcptw->tw_rcv_nxt))
 61		WRITE_ONCE(ao->rcv_sne, ao->rcv_sne + 1);
 62#endif
 63	tcptw->tw_rcv_nxt = seq;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 64}
 65
 66/*
 67 * * Main purpose of TIME-WAIT state is to close connection gracefully,
 68 *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
 69 *   (and, probably, tail of data) and one or more our ACKs are lost.
 70 * * What is TIME-WAIT timeout? It is associated with maximal packet
 71 *   lifetime in the internet, which results in wrong conclusion, that
 72 *   it is set to catch "old duplicate segments" wandering out of their path.
 73 *   It is not quite correct. This timeout is calculated so that it exceeds
 74 *   maximal retransmission timeout enough to allow to lose one (or more)
 75 *   segments sent by peer and our ACKs. This time may be calculated from RTO.
 76 * * When TIME-WAIT socket receives RST, it means that another end
 77 *   finally closed and we are allowed to kill TIME-WAIT too.
 78 * * Second purpose of TIME-WAIT is catching old duplicate segments.
 79 *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
 80 *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
 81 * * If we invented some more clever way to catch duplicates
 82 *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
 83 *
 84 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
 85 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
 86 * from the very beginning.
 87 *
 88 * NOTE. With recycling (and later with fin-wait-2) TW bucket
 89 * is _not_ stateless. It means, that strictly speaking we must
 90 * spinlock it. I do not want! Well, probability of misbehaviour
 91 * is ridiculously low and, seems, we could use some mb() tricks
 92 * to avoid misread sequence numbers, states etc.  --ANK
 93 *
 94 * We don't need to initialize tmp_out.sack_ok as we don't use the results
 95 */
 96enum tcp_tw_status
 97tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
 98			   const struct tcphdr *th)
 99{
100	struct tcp_options_received tmp_opt;
 
101	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
102	bool paws_reject = false;
103
104	tmp_opt.saw_tstamp = 0;
105	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
106		tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
107
108		if (tmp_opt.saw_tstamp) {
109			if (tmp_opt.rcv_tsecr)
110				tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
111			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
112			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
113			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
114		}
115	}
116
117	if (tw->tw_substate == TCP_FIN_WAIT2) {
118		/* Just repeat all the checks of tcp_rcv_state_process() */
119
120		/* Out of window, send ACK */
121		if (paws_reject ||
122		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
123				   tcptw->tw_rcv_nxt,
124				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
125			return tcp_timewait_check_oow_rate_limit(
126				tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
127
128		if (th->rst)
129			goto kill;
130
131		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
132			return TCP_TW_RST;
133
134		/* Dup ACK? */
135		if (!th->ack ||
136		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
137		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
138			inet_twsk_put(tw);
139			return TCP_TW_SUCCESS;
140		}
141
142		/* New data or FIN. If new data arrive after half-duplex close,
143		 * reset.
144		 */
145		if (!th->fin ||
146		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
 
 
 
147			return TCP_TW_RST;
 
148
149		/* FIN arrived, enter true time-wait state. */
150		tw->tw_substate	  = TCP_TIME_WAIT;
151		twsk_rcv_nxt_update(tcptw, TCP_SKB_CB(skb)->end_seq);
152
153		if (tmp_opt.saw_tstamp) {
154			tcptw->tw_ts_recent_stamp = ktime_get_seconds();
155			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
156		}
157
158		inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
 
 
 
 
 
 
 
159		return TCP_TW_ACK;
160	}
161
162	/*
163	 *	Now real TIME-WAIT state.
164	 *
165	 *	RFC 1122:
166	 *	"When a connection is [...] on TIME-WAIT state [...]
167	 *	[a TCP] MAY accept a new SYN from the remote TCP to
168	 *	reopen the connection directly, if it:
169	 *
170	 *	(1)  assigns its initial sequence number for the new
171	 *	connection to be larger than the largest sequence
172	 *	number it used on the previous connection incarnation,
173	 *	and
174	 *
175	 *	(2)  returns to TIME-WAIT state if the SYN turns out
176	 *	to be an old duplicate".
177	 */
178
179	if (!paws_reject &&
180	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
181	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
182		/* In window segment, it may be only reset or bare ack. */
183
184		if (th->rst) {
185			/* This is TIME_WAIT assassination, in two flavors.
186			 * Oh well... nobody has a sufficient solution to this
187			 * protocol bug yet.
188			 */
189			if (!READ_ONCE(twsk_net(tw)->ipv4.sysctl_tcp_rfc1337)) {
190kill:
191				inet_twsk_deschedule_put(tw);
 
192				return TCP_TW_SUCCESS;
193			}
194		} else {
195			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
196		}
 
 
197
198		if (tmp_opt.saw_tstamp) {
199			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
200			tcptw->tw_ts_recent_stamp = ktime_get_seconds();
201		}
202
203		inet_twsk_put(tw);
204		return TCP_TW_SUCCESS;
205	}
206
207	/* Out of window segment.
208
209	   All the segments are ACKed immediately.
210
211	   The only exception is new SYN. We accept it, if it is
212	   not old duplicate and we are not in danger to be killed
213	   by delayed old duplicates. RFC check is that it has
214	   newer sequence number works at rates <40Mbit/sec.
215	   However, if paws works, it is reliable AND even more,
216	   we even may relax silly seq space cutoff.
217
218	   RED-PEN: we violate main RFC requirement, if this SYN will appear
219	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
220	   we must return socket to time-wait state. It is not good,
221	   but not fatal yet.
222	 */
223
224	if (th->syn && !th->rst && !th->ack && !paws_reject &&
225	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
226	     (tmp_opt.saw_tstamp &&
227	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
228		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
229		if (isn == 0)
230			isn++;
231		TCP_SKB_CB(skb)->tcp_tw_isn = isn;
232		return TCP_TW_SYN;
233	}
234
235	if (paws_reject)
236		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
237
238	if (!th->rst) {
239		/* In this case we must reset the TIMEWAIT timer.
240		 *
241		 * If it is ACKless SYN it may be both old duplicate
242		 * and new good SYN with random sequence number <rcv_nxt.
243		 * Do not reschedule in the last case.
244		 */
245		if (paws_reject || th->ack)
246			inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
 
247
248		return tcp_timewait_check_oow_rate_limit(
249			tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
 
 
250	}
251	inet_twsk_put(tw);
252	return TCP_TW_SUCCESS;
253}
254EXPORT_SYMBOL(tcp_timewait_state_process);
255
256static void tcp_time_wait_init(struct sock *sk, struct tcp_timewait_sock *tcptw)
257{
258#ifdef CONFIG_TCP_MD5SIG
259	const struct tcp_sock *tp = tcp_sk(sk);
260	struct tcp_md5sig_key *key;
261
262	/*
263	 * The timewait bucket does not have the key DB from the
264	 * sock structure. We just make a quick copy of the
265	 * md5 key being used (if indeed we are using one)
266	 * so the timewait ack generating code has the key.
267	 */
268	tcptw->tw_md5_key = NULL;
269	if (!static_branch_unlikely(&tcp_md5_needed.key))
270		return;
271
272	key = tp->af_specific->md5_lookup(sk, sk);
273	if (key) {
274		tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
275		if (!tcptw->tw_md5_key)
276			return;
277		if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key))
278			goto out_free;
279		tcp_md5_add_sigpool();
280	}
281	return;
282out_free:
283	WARN_ON_ONCE(1);
284	kfree(tcptw->tw_md5_key);
285	tcptw->tw_md5_key = NULL;
286#endif
287}
288
289/*
290 * Move a socket to time-wait or dead fin-wait-2 state.
291 */
292void tcp_time_wait(struct sock *sk, int state, int timeo)
293{
 
294	const struct inet_connection_sock *icsk = inet_csk(sk);
295	struct tcp_sock *tp = tcp_sk(sk);
296	struct net *net = sock_net(sk);
297	struct inet_timewait_sock *tw;
 
 
298
299	tw = inet_twsk_alloc(sk, &net->ipv4.tcp_death_row, state);
 
300
301	if (tw) {
302		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
303		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
304
305		tw->tw_transparent	= inet_test_bit(TRANSPARENT, sk);
306		tw->tw_mark		= sk->sk_mark;
307		tw->tw_priority		= READ_ONCE(sk->sk_priority);
308		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
309		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
310		tcptw->tw_snd_nxt	= tp->snd_nxt;
311		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
312		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
313		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
314		tcptw->tw_ts_offset	= tp->tsoffset;
315		tw->tw_usec_ts		= tp->tcp_usec_ts;
316		tcptw->tw_last_oow_ack_time = 0;
317		tcptw->tw_tx_delay	= tp->tcp_tx_delay;
318		tw->tw_txhash		= sk->sk_txhash;
319#if IS_ENABLED(CONFIG_IPV6)
320		if (tw->tw_family == PF_INET6) {
321			struct ipv6_pinfo *np = inet6_sk(sk);
 
322
323			tw->tw_v6_daddr = sk->sk_v6_daddr;
324			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
325			tw->tw_tclass = np->tclass;
326			tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
327			tw->tw_ipv6only = sk->sk_ipv6only;
328		}
329#endif
330
331		tcp_time_wait_init(sk, tcptw);
332		tcp_ao_time_wait(tcptw, tp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
333
334		/* Get the TIME_WAIT timeout firing. */
335		if (timeo < rto)
336			timeo = rto;
337
338		if (state == TCP_TIME_WAIT)
339			timeo = TCP_TIMEWAIT_LEN;
 
 
 
 
 
340
341		/* tw_timer is pinned, so we need to make sure BH are disabled
342		 * in following section, otherwise timer handler could run before
343		 * we complete the initialization.
344		 */
345		local_bh_disable();
346		inet_twsk_schedule(tw, timeo);
347		/* Linkage updates.
348		 * Note that access to tw after this point is illegal.
349		 */
350		inet_twsk_hashdance(tw, sk, net->ipv4.tcp_death_row.hashinfo);
351		local_bh_enable();
352	} else {
353		/* Sorry, if we're out of memory, just CLOSE this
354		 * socket up.  We've got bigger problems than
355		 * non-graceful socket closings.
356		 */
357		NET_INC_STATS(net, LINUX_MIB_TCPTIMEWAITOVERFLOW);
358	}
359
360	tcp_update_metrics(sk);
361	tcp_done(sk);
362}
363EXPORT_SYMBOL(tcp_time_wait);
364
365#ifdef CONFIG_TCP_MD5SIG
366static void tcp_md5_twsk_free_rcu(struct rcu_head *head)
367{
368	struct tcp_md5sig_key *key;
369
370	key = container_of(head, struct tcp_md5sig_key, rcu);
371	kfree(key);
372	static_branch_slow_dec_deferred(&tcp_md5_needed);
373	tcp_md5_release_sigpool();
374}
375#endif
376
377void tcp_twsk_destructor(struct sock *sk)
378{
379#ifdef CONFIG_TCP_MD5SIG
380	if (static_branch_unlikely(&tcp_md5_needed.key)) {
381		struct tcp_timewait_sock *twsk = tcp_twsk(sk);
382
383		if (twsk->tw_md5_key)
384			call_rcu(&twsk->tw_md5_key->rcu, tcp_md5_twsk_free_rcu);
385	}
386#endif
387	tcp_ao_destroy_sock(sk, true);
388}
389EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
390
391void tcp_twsk_purge(struct list_head *net_exit_list, int family)
392{
393	bool purged_once = false;
394	struct net *net;
395
396	list_for_each_entry(net, net_exit_list, exit_list) {
397		if (net->ipv4.tcp_death_row.hashinfo->pernet) {
398			/* Even if tw_refcount == 1, we must clean up kernel reqsk */
399			inet_twsk_purge(net->ipv4.tcp_death_row.hashinfo, family);
400		} else if (!purged_once) {
401			inet_twsk_purge(&tcp_hashinfo, family);
402			purged_once = true;
403		}
404	}
405}
406EXPORT_SYMBOL_GPL(tcp_twsk_purge);
407
408/* Warning : This function is called without sk_listener being locked.
409 * Be sure to read socket fields once, as their value could change under us.
410 */
411void tcp_openreq_init_rwin(struct request_sock *req,
412			   const struct sock *sk_listener,
413			   const struct dst_entry *dst)
414{
415	struct inet_request_sock *ireq = inet_rsk(req);
416	const struct tcp_sock *tp = tcp_sk(sk_listener);
417	int full_space = tcp_full_space(sk_listener);
418	u32 window_clamp;
419	__u8 rcv_wscale;
420	u32 rcv_wnd;
421	int mss;
422
423	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
424	window_clamp = READ_ONCE(tp->window_clamp);
425	/* Set this up on the first call only */
426	req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
427
428	/* limit the window selection if the user enforce a smaller rx buffer */
429	if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
430	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
431		req->rsk_window_clamp = full_space;
432
433	rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
434	if (rcv_wnd == 0)
435		rcv_wnd = dst_metric(dst, RTAX_INITRWND);
436	else if (full_space < rcv_wnd * mss)
437		full_space = rcv_wnd * mss;
438
439	/* tcp_full_space because it is guaranteed to be the first packet */
440	tcp_select_initial_window(sk_listener, full_space,
441		mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
442		&req->rsk_rcv_wnd,
443		&req->rsk_window_clamp,
444		ireq->wscale_ok,
445		&rcv_wscale,
446		rcv_wnd);
447	ireq->rcv_wscale = rcv_wscale;
448}
449EXPORT_SYMBOL(tcp_openreq_init_rwin);
450
451static void tcp_ecn_openreq_child(struct tcp_sock *tp,
452				  const struct request_sock *req)
453{
454	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
455}
456
457void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
458{
459	struct inet_connection_sock *icsk = inet_csk(sk);
460	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
461	bool ca_got_dst = false;
462
463	if (ca_key != TCP_CA_UNSPEC) {
464		const struct tcp_congestion_ops *ca;
465
466		rcu_read_lock();
467		ca = tcp_ca_find_key(ca_key);
468		if (likely(ca && bpf_try_module_get(ca, ca->owner))) {
469			icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
470			icsk->icsk_ca_ops = ca;
471			ca_got_dst = true;
472		}
473		rcu_read_unlock();
474	}
475
476	/* If no valid choice made yet, assign current system default ca. */
477	if (!ca_got_dst &&
478	    (!icsk->icsk_ca_setsockopt ||
479	     !bpf_try_module_get(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner)))
480		tcp_assign_congestion_control(sk);
481
482	tcp_set_ca_state(sk, TCP_CA_Open);
483}
484EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
485
486static void smc_check_reset_syn_req(const struct tcp_sock *oldtp,
487				    struct request_sock *req,
488				    struct tcp_sock *newtp)
489{
490#if IS_ENABLED(CONFIG_SMC)
491	struct inet_request_sock *ireq;
492
493	if (static_branch_unlikely(&tcp_have_smc)) {
494		ireq = inet_rsk(req);
495		if (oldtp->syn_smc && !ireq->smc_ok)
496			newtp->syn_smc = 0;
497	}
498#endif
499}
500
501/* This is not only more efficient than what we used to do, it eliminates
502 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
503 *
504 * Actually, we could lots of memory writes here. tp of listening
505 * socket contains all necessary default parameters.
506 */
507struct sock *tcp_create_openreq_child(const struct sock *sk,
508				      struct request_sock *req,
509				      struct sk_buff *skb)
510{
511	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
512	const struct inet_request_sock *ireq = inet_rsk(req);
513	struct tcp_request_sock *treq = tcp_rsk(req);
514	struct inet_connection_sock *newicsk;
515	const struct tcp_sock *oldtp;
516	struct tcp_sock *newtp;
517	u32 seq;
518#ifdef CONFIG_TCP_AO
519	struct tcp_ao_key *ao_key;
520#endif
521
522	if (!newsk)
523		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
524
525	newicsk = inet_csk(newsk);
526	newtp = tcp_sk(newsk);
527	oldtp = tcp_sk(sk);
528
529	smc_check_reset_syn_req(oldtp, req, newtp);
530
531	/* Now setup tcp_sock */
532	newtp->pred_flags = 0;
533
534	seq = treq->rcv_isn + 1;
535	newtp->rcv_wup = seq;
536	WRITE_ONCE(newtp->copied_seq, seq);
537	WRITE_ONCE(newtp->rcv_nxt, seq);
538	newtp->segs_in = 1;
539
540	seq = treq->snt_isn + 1;
541	newtp->snd_sml = newtp->snd_una = seq;
542	WRITE_ONCE(newtp->snd_nxt, seq);
543	newtp->snd_up = seq;
544
545	INIT_LIST_HEAD(&newtp->tsq_node);
546	INIT_LIST_HEAD(&newtp->tsorted_sent_queue);
547
548	tcp_init_wl(newtp, treq->rcv_isn);
549
550	minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
551	newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
552
553	newtp->lsndtime = tcp_jiffies32;
554	newsk->sk_txhash = READ_ONCE(treq->txhash);
555	newtp->total_retrans = req->num_retrans;
556
557	tcp_init_xmit_timers(newsk);
558	WRITE_ONCE(newtp->write_seq, newtp->pushed_seq = treq->snt_isn + 1);
559
560	if (sock_flag(newsk, SOCK_KEEPOPEN))
561		inet_csk_reset_keepalive_timer(newsk,
562					       keepalive_time_when(newtp));
563
564	newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
565	newtp->rx_opt.sack_ok = ireq->sack_ok;
566	newtp->window_clamp = req->rsk_window_clamp;
567	newtp->rcv_ssthresh = req->rsk_rcv_wnd;
568	newtp->rcv_wnd = req->rsk_rcv_wnd;
569	newtp->rx_opt.wscale_ok = ireq->wscale_ok;
570	if (newtp->rx_opt.wscale_ok) {
571		newtp->rx_opt.snd_wscale = ireq->snd_wscale;
572		newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
573	} else {
574		newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
575		newtp->window_clamp = min(newtp->window_clamp, 65535U);
576	}
577	newtp->snd_wnd = ntohs(tcp_hdr(skb)->window) << newtp->rx_opt.snd_wscale;
578	newtp->max_window = newtp->snd_wnd;
579
580	if (newtp->rx_opt.tstamp_ok) {
581		newtp->tcp_usec_ts = treq->req_usec_ts;
582		newtp->rx_opt.ts_recent = READ_ONCE(req->ts_recent);
583		newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
584		newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
585	} else {
586		newtp->tcp_usec_ts = 0;
587		newtp->rx_opt.ts_recent_stamp = 0;
588		newtp->tcp_header_len = sizeof(struct tcphdr);
589	}
590	if (req->num_timeout) {
591		newtp->total_rto = req->num_timeout;
592		newtp->undo_marker = treq->snt_isn;
593		if (newtp->tcp_usec_ts) {
594			newtp->retrans_stamp = treq->snt_synack;
595			newtp->total_rto_time = (u32)(tcp_clock_us() -
596						      newtp->retrans_stamp) / USEC_PER_MSEC;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
597		} else {
598			newtp->retrans_stamp = div_u64(treq->snt_synack,
599						       USEC_PER_SEC / TCP_TS_HZ);
600			newtp->total_rto_time = tcp_clock_ms() -
601						newtp->retrans_stamp;
 
 
 
 
 
 
 
 
 
 
602		}
603		newtp->total_rto_recoveries = 1;
604	}
605	newtp->tsoffset = treq->ts_off;
606#ifdef CONFIG_TCP_MD5SIG
607	newtp->md5sig_info = NULL;	/*XXX*/
 
 
608#endif
609#ifdef CONFIG_TCP_AO
610	newtp->ao_info = NULL;
611	ao_key = treq->af_specific->ao_lookup(sk, req,
612				tcp_rsk(req)->ao_keyid, -1);
613	if (ao_key)
614		newtp->tcp_header_len += tcp_ao_len_aligned(ao_key);
615 #endif
616	if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
617		newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
618	newtp->rx_opt.mss_clamp = req->mss;
619	tcp_ecn_openreq_child(newtp, req);
620	newtp->fastopen_req = NULL;
621	RCU_INIT_POINTER(newtp->fastopen_rsk, NULL);
622
623	newtp->bpf_chg_cc_inprogress = 0;
624	tcp_bpf_clone(sk, newsk);
625
626	__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
627
 
 
628	return newsk;
629}
630EXPORT_SYMBOL(tcp_create_openreq_child);
631
632/*
633 * Process an incoming packet for SYN_RECV sockets represented as a
634 * request_sock. Normally sk is the listener socket but for TFO it
635 * points to the child socket.
636 *
637 * XXX (TFO) - The current impl contains a special check for ack
638 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
639 *
640 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
641 *
642 * Note: If @fastopen is true, this can be called from process context.
643 *       Otherwise, this is from BH context.
644 */
645
646struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
647			   struct request_sock *req,
648			   bool fastopen, bool *req_stolen)
649{
650	struct tcp_options_received tmp_opt;
 
651	struct sock *child;
652	const struct tcphdr *th = tcp_hdr(skb);
653	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
654	bool paws_reject = false;
655	bool own_req;
656
657	tmp_opt.saw_tstamp = 0;
658	if (th->doff > (sizeof(struct tcphdr)>>2)) {
659		tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
660
661		if (tmp_opt.saw_tstamp) {
662			tmp_opt.ts_recent = READ_ONCE(req->ts_recent);
663			if (tmp_opt.rcv_tsecr)
664				tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
665			/* We do not store true stamp, but it is not required,
666			 * it can be estimated (approximately)
667			 * from another data.
668			 */
669			tmp_opt.ts_recent_stamp = ktime_get_seconds() - reqsk_timeout(req, TCP_RTO_MAX) / HZ;
670			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
671		}
672	}
673
674	/* Check for pure retransmitted SYN. */
675	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
676	    flg == TCP_FLAG_SYN &&
677	    !paws_reject) {
678		/*
679		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
680		 * this case on figure 6 and figure 8, but formal
681		 * protocol description says NOTHING.
682		 * To be more exact, it says that we should send ACK,
683		 * because this segment (at least, if it has no data)
684		 * is out of window.
685		 *
686		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
687		 *  describe SYN-RECV state. All the description
688		 *  is wrong, we cannot believe to it and should
689		 *  rely only on common sense and implementation
690		 *  experience.
691		 *
692		 * Enforce "SYN-ACK" according to figure 8, figure 6
693		 * of RFC793, fixed by RFC1122.
694		 *
695		 * Note that even if there is new data in the SYN packet
696		 * they will be thrown away too.
697		 *
698		 * Reset timer after retransmitting SYNACK, similar to
699		 * the idea of fast retransmit in recovery.
700		 */
701		if (!tcp_oow_rate_limited(sock_net(sk), skb,
702					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
703					  &tcp_rsk(req)->last_oow_ack_time) &&
704
705		    !inet_rtx_syn_ack(sk, req)) {
706			unsigned long expires = jiffies;
707
708			expires += reqsk_timeout(req, TCP_RTO_MAX);
709			if (!fastopen)
710				mod_timer_pending(&req->rsk_timer, expires);
711			else
712				req->rsk_timer.expires = expires;
713		}
714		return NULL;
715	}
716
717	/* Further reproduces section "SEGMENT ARRIVES"
718	   for state SYN-RECEIVED of RFC793.
719	   It is broken, however, it does not work only
720	   when SYNs are crossed.
721
722	   You would think that SYN crossing is impossible here, since
723	   we should have a SYN_SENT socket (from connect()) on our end,
724	   but this is not true if the crossed SYNs were sent to both
725	   ends by a malicious third party.  We must defend against this,
726	   and to do that we first verify the ACK (as per RFC793, page
727	   36) and reset if it is invalid.  Is this a true full defense?
728	   To convince ourselves, let us consider a way in which the ACK
729	   test can still pass in this 'malicious crossed SYNs' case.
730	   Malicious sender sends identical SYNs (and thus identical sequence
731	   numbers) to both A and B:
732
733		A: gets SYN, seq=7
734		B: gets SYN, seq=7
735
736	   By our good fortune, both A and B select the same initial
737	   send sequence number of seven :-)
738
739		A: sends SYN|ACK, seq=7, ack_seq=8
740		B: sends SYN|ACK, seq=7, ack_seq=8
741
742	   So we are now A eating this SYN|ACK, ACK test passes.  So
743	   does sequence test, SYN is truncated, and thus we consider
744	   it a bare ACK.
745
746	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
747	   bare ACK.  Otherwise, we create an established connection.  Both
748	   ends (listening sockets) accept the new incoming connection and try
749	   to talk to each other. 8-)
750
751	   Note: This case is both harmless, and rare.  Possibility is about the
752	   same as us discovering intelligent life on another plant tomorrow.
753
754	   But generally, we should (RFC lies!) to accept ACK
755	   from SYNACK both here and in tcp_rcv_state_process().
756	   tcp_rcv_state_process() does not, hence, we do not too.
757
758	   Note that the case is absolutely generic:
759	   we cannot optimize anything here without
760	   violating protocol. All the checks must be made
761	   before attempt to create socket.
762	 */
763
764	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
765	 *                  and the incoming segment acknowledges something not yet
766	 *                  sent (the segment carries an unacceptable ACK) ...
767	 *                  a reset is sent."
768	 *
769	 * Invalid ACK: reset will be sent by listening socket.
770	 * Note that the ACK validity check for a Fast Open socket is done
771	 * elsewhere and is checked directly against the child socket rather
772	 * than req because user data may have been sent out.
773	 */
774	if ((flg & TCP_FLAG_ACK) && !fastopen &&
775	    (TCP_SKB_CB(skb)->ack_seq !=
776	     tcp_rsk(req)->snt_isn + 1))
777		return sk;
778
779	/* Also, it would be not so bad idea to check rcv_tsecr, which
780	 * is essentially ACK extension and too early or too late values
781	 * should cause reset in unsynchronized states.
782	 */
783
784	/* RFC793: "first check sequence number". */
785
786	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
787					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
788		/* Out of window: send ACK and drop. */
789		if (!(flg & TCP_FLAG_RST) &&
790		    !tcp_oow_rate_limited(sock_net(sk), skb,
791					  LINUX_MIB_TCPACKSKIPPEDSYNRECV,
792					  &tcp_rsk(req)->last_oow_ack_time))
793			req->rsk_ops->send_ack(sk, skb, req);
794		if (paws_reject)
795			NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
796		return NULL;
797	}
798
799	/* In sequence, PAWS is OK. */
800
801	/* TODO: We probably should defer ts_recent change once
802	 * we take ownership of @req.
803	 */
804	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
805		WRITE_ONCE(req->ts_recent, tmp_opt.rcv_tsval);
806
807	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
808		/* Truncate SYN, it is out of window starting
809		   at tcp_rsk(req)->rcv_isn + 1. */
810		flg &= ~TCP_FLAG_SYN;
811	}
812
813	/* RFC793: "second check the RST bit" and
814	 *	   "fourth, check the SYN bit"
815	 */
816	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
817		TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
818		goto embryonic_reset;
819	}
820
821	/* ACK sequence verified above, just make sure ACK is
822	 * set.  If ACK not set, just silently drop the packet.
823	 *
824	 * XXX (TFO) - if we ever allow "data after SYN", the
825	 * following check needs to be removed.
826	 */
827	if (!(flg & TCP_FLAG_ACK))
828		return NULL;
829
830	/* For Fast Open no more processing is needed (sk is the
831	 * child socket).
832	 */
833	if (fastopen)
834		return sk;
835
836	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
837	if (req->num_timeout < READ_ONCE(inet_csk(sk)->icsk_accept_queue.rskq_defer_accept) &&
838	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
839		inet_rsk(req)->acked = 1;
840		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
841		return NULL;
842	}
 
 
 
 
843
844	/* OK, ACK is valid, create big socket and
845	 * feed this segment to it. It will repeat all
846	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
847	 * ESTABLISHED STATE. If it will be dropped after
848	 * socket is created, wait for troubles.
849	 */
850	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
851							 req, &own_req);
852	if (!child)
853		goto listen_overflow;
854
855	if (own_req && rsk_drop_req(req)) {
856		reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
857		inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
858		return child;
859	}
860
861	sock_rps_save_rxhash(child, skb);
862	tcp_synack_rtt_meas(child, req);
863	*req_stolen = !own_req;
864	return inet_csk_complete_hashdance(sk, child, req, own_req);
865
866listen_overflow:
867	if (sk != req->rsk_listener)
868		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
869
870	if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow)) {
871		inet_rsk(req)->acked = 1;
872		return NULL;
873	}
874
875embryonic_reset:
876	if (!(flg & TCP_FLAG_RST)) {
877		/* Received a bad SYN pkt - for TFO We try not to reset
878		 * the local connection unless it's really necessary to
879		 * avoid becoming vulnerable to outside attack aiming at
880		 * resetting legit local connections.
881		 */
882		req->rsk_ops->send_reset(sk, skb);
883	} else if (fastopen) { /* received a valid RST pkt */
884		reqsk_fastopen_remove(sk, req, true);
885		tcp_reset(sk, skb);
886	}
887	if (!fastopen) {
888		bool unlinked = inet_csk_reqsk_queue_drop(sk, req);
889
890		if (unlinked)
891			__NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
892		*req_stolen = !unlinked;
893	}
894	return NULL;
895}
896EXPORT_SYMBOL(tcp_check_req);
897
898/*
899 * Queue segment on the new socket if the new socket is active,
900 * otherwise we just shortcircuit this and continue with
901 * the new socket.
902 *
903 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
904 * when entering. But other states are possible due to a race condition
905 * where after __inet_lookup_established() fails but before the listener
906 * locked is obtained, other packets cause the same connection to
907 * be created.
908 */
909
910enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
911				       struct sk_buff *skb)
912	__releases(&((child)->sk_lock.slock))
913{
914	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
915	int state = child->sk_state;
916
917	/* record sk_napi_id and sk_rx_queue_mapping of child. */
918	sk_mark_napi_id_set(child, skb);
919
920	tcp_segs_in(tcp_sk(child), skb);
921	if (!sock_owned_by_user(child)) {
922		reason = tcp_rcv_state_process(child, skb);
 
923		/* Wakeup parent, send SIGIO */
924		if (state == TCP_SYN_RECV && child->sk_state != state)
925			parent->sk_data_ready(parent);
926	} else {
927		/* Alas, it is possible again, because we do lookup
928		 * in main socket hash table and lock on listening
929		 * socket does not protect us more.
930		 */
931		__sk_add_backlog(child, skb);
932	}
933
934	bh_unlock_sock(child);
935	sock_put(child);
936	return reason;
937}
938EXPORT_SYMBOL(tcp_child_process);
v3.1
 
  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
 30int sysctl_tcp_syncookies __read_mostly = 1;
 31EXPORT_SYMBOL(sysctl_tcp_syncookies);
 32
 33int sysctl_tcp_abort_on_overflow __read_mostly;
 34
 35struct 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};
 50EXPORT_SYMBOL_GPL(tcp_death_row);
 51
 52/* VJ's idea. Save last timestamp seen from this destination
 53 * and hold it at least for normal timewait interval to use for duplicate
 54 * segment detection in subsequent connections, before they enter synchronized
 55 * state.
 56 */
 57
 58static int tcp_remember_stamp(struct sock *sk)
 
 
 59{
 60	const struct inet_connection_sock *icsk = inet_csk(sk);
 61	struct tcp_sock *tp = tcp_sk(sk);
 62	struct inet_peer *peer;
 63	bool release_it;
 64
 65	peer = icsk->icsk_af_ops->get_peer(sk, &release_it);
 66	if (peer) {
 67		if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
 68		    ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
 69		     peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
 70			peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
 71			peer->tcp_ts = tp->rx_opt.ts_recent;
 72		}
 73		if (release_it)
 74			inet_putpeer(peer);
 75		return 1;
 76	}
 77
 78	return 0;
 
 
 79}
 80
 81static int tcp_tw_remember_stamp(struct inet_timewait_sock *tw)
 82{
 83	struct sock *sk = (struct sock *) tw;
 84	struct inet_peer *peer;
 85
 86	peer = twsk_getpeer(sk);
 87	if (peer) {
 88		const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 89
 90		if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
 91		    ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
 92		     peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
 93			peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
 94			peer->tcp_ts	   = tcptw->tw_ts_recent;
 95		}
 96		inet_putpeer(peer);
 97		return 1;
 98	}
 99	return 0;
100}
101
102static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
103{
104	if (seq == s_win)
105		return 1;
106	if (after(end_seq, s_win) && before(seq, e_win))
107		return 1;
108	return seq == e_win && seq == end_seq;
109}
110
111/*
112 * * Main purpose of TIME-WAIT state is to close connection gracefully,
113 *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
114 *   (and, probably, tail of data) and one or more our ACKs are lost.
115 * * What is TIME-WAIT timeout? It is associated with maximal packet
116 *   lifetime in the internet, which results in wrong conclusion, that
117 *   it is set to catch "old duplicate segments" wandering out of their path.
118 *   It is not quite correct. This timeout is calculated so that it exceeds
119 *   maximal retransmission timeout enough to allow to lose one (or more)
120 *   segments sent by peer and our ACKs. This time may be calculated from RTO.
121 * * When TIME-WAIT socket receives RST, it means that another end
122 *   finally closed and we are allowed to kill TIME-WAIT too.
123 * * Second purpose of TIME-WAIT is catching old duplicate segments.
124 *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
125 *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
126 * * If we invented some more clever way to catch duplicates
127 *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
128 *
129 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
130 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
131 * from the very beginning.
132 *
133 * NOTE. With recycling (and later with fin-wait-2) TW bucket
134 * is _not_ stateless. It means, that strictly speaking we must
135 * spinlock it. I do not want! Well, probability of misbehaviour
136 * is ridiculously low and, seems, we could use some mb() tricks
137 * to avoid misread sequence numbers, states etc.  --ANK
 
 
138 */
139enum tcp_tw_status
140tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
141			   const struct tcphdr *th)
142{
143	struct tcp_options_received tmp_opt;
144	u8 *hash_location;
145	struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
146	int paws_reject = 0;
147
148	tmp_opt.saw_tstamp = 0;
149	if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
150		tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
151
152		if (tmp_opt.saw_tstamp) {
 
 
153			tmp_opt.ts_recent	= tcptw->tw_ts_recent;
154			tmp_opt.ts_recent_stamp	= tcptw->tw_ts_recent_stamp;
155			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
156		}
157	}
158
159	if (tw->tw_substate == TCP_FIN_WAIT2) {
160		/* Just repeat all the checks of tcp_rcv_state_process() */
161
162		/* Out of window, send ACK */
163		if (paws_reject ||
164		    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
165				   tcptw->tw_rcv_nxt,
166				   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
167			return TCP_TW_ACK;
 
168
169		if (th->rst)
170			goto kill;
171
172		if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
173			goto kill_with_rst;
174
175		/* Dup ACK? */
176		if (!th->ack ||
177		    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
178		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
179			inet_twsk_put(tw);
180			return TCP_TW_SUCCESS;
181		}
182
183		/* New data or FIN. If new data arrive after half-duplex close,
184		 * reset.
185		 */
186		if (!th->fin ||
187		    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
188kill_with_rst:
189			inet_twsk_deschedule(tw, &tcp_death_row);
190			inet_twsk_put(tw);
191			return TCP_TW_RST;
192		}
193
194		/* FIN arrived, enter true time-wait state. */
195		tw->tw_substate	  = TCP_TIME_WAIT;
196		tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
 
197		if (tmp_opt.saw_tstamp) {
198			tcptw->tw_ts_recent_stamp = get_seconds();
199			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
200		}
201
202		if (tcp_death_row.sysctl_tw_recycle &&
203		    tcptw->tw_ts_recent_stamp &&
204		    tcp_tw_remember_stamp(tw))
205			inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
206					   TCP_TIMEWAIT_LEN);
207		else
208			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
209					   TCP_TIMEWAIT_LEN);
210		return TCP_TW_ACK;
211	}
212
213	/*
214	 *	Now real TIME-WAIT state.
215	 *
216	 *	RFC 1122:
217	 *	"When a connection is [...] on TIME-WAIT state [...]
218	 *	[a TCP] MAY accept a new SYN from the remote TCP to
219	 *	reopen the connection directly, if it:
220	 *
221	 *	(1)  assigns its initial sequence number for the new
222	 *	connection to be larger than the largest sequence
223	 *	number it used on the previous connection incarnation,
224	 *	and
225	 *
226	 *	(2)  returns to TIME-WAIT state if the SYN turns out
227	 *	to be an old duplicate".
228	 */
229
230	if (!paws_reject &&
231	    (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
232	     (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
233		/* In window segment, it may be only reset or bare ack. */
234
235		if (th->rst) {
236			/* This is TIME_WAIT assassination, in two flavors.
237			 * Oh well... nobody has a sufficient solution to this
238			 * protocol bug yet.
239			 */
240			if (sysctl_tcp_rfc1337 == 0) {
241kill:
242				inet_twsk_deschedule(tw, &tcp_death_row);
243				inet_twsk_put(tw);
244				return TCP_TW_SUCCESS;
245			}
 
 
246		}
247		inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
248				   TCP_TIMEWAIT_LEN);
249
250		if (tmp_opt.saw_tstamp) {
251			tcptw->tw_ts_recent	  = tmp_opt.rcv_tsval;
252			tcptw->tw_ts_recent_stamp = get_seconds();
253		}
254
255		inet_twsk_put(tw);
256		return TCP_TW_SUCCESS;
257	}
258
259	/* Out of window segment.
260
261	   All the segments are ACKed immediately.
262
263	   The only exception is new SYN. We accept it, if it is
264	   not old duplicate and we are not in danger to be killed
265	   by delayed old duplicates. RFC check is that it has
266	   newer sequence number works at rates <40Mbit/sec.
267	   However, if paws works, it is reliable AND even more,
268	   we even may relax silly seq space cutoff.
269
270	   RED-PEN: we violate main RFC requirement, if this SYN will appear
271	   old duplicate (i.e. we receive RST in reply to SYN-ACK),
272	   we must return socket to time-wait state. It is not good,
273	   but not fatal yet.
274	 */
275
276	if (th->syn && !th->rst && !th->ack && !paws_reject &&
277	    (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
278	     (tmp_opt.saw_tstamp &&
279	      (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
280		u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
281		if (isn == 0)
282			isn++;
283		TCP_SKB_CB(skb)->when = isn;
284		return TCP_TW_SYN;
285	}
286
287	if (paws_reject)
288		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
289
290	if (!th->rst) {
291		/* In this case we must reset the TIMEWAIT timer.
292		 *
293		 * If it is ACKless SYN it may be both old duplicate
294		 * and new good SYN with random sequence number <rcv_nxt.
295		 * Do not reschedule in the last case.
296		 */
297		if (paws_reject || th->ack)
298			inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
299					   TCP_TIMEWAIT_LEN);
300
301		/* Send ACK. Note, we do not put the bucket,
302		 * it will be released by caller.
303		 */
304		return TCP_TW_ACK;
305	}
306	inet_twsk_put(tw);
307	return TCP_TW_SUCCESS;
308}
309EXPORT_SYMBOL(tcp_timewait_state_process);
310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
311/*
312 * Move a socket to time-wait or dead fin-wait-2 state.
313 */
314void tcp_time_wait(struct sock *sk, int state, int timeo)
315{
316	struct inet_timewait_sock *tw = NULL;
317	const struct inet_connection_sock *icsk = inet_csk(sk);
318	const struct tcp_sock *tp = tcp_sk(sk);
319	int recycle_ok = 0;
320
321	if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
322		recycle_ok = tcp_remember_stamp(sk);
323
324	if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
325		tw = inet_twsk_alloc(sk, state);
326
327	if (tw != NULL) {
328		struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
329		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
330
331		tw->tw_transparent	= inet_sk(sk)->transparent;
 
 
332		tw->tw_rcv_wscale	= tp->rx_opt.rcv_wscale;
333		tcptw->tw_rcv_nxt	= tp->rcv_nxt;
334		tcptw->tw_snd_nxt	= tp->snd_nxt;
335		tcptw->tw_rcv_wnd	= tcp_receive_window(tp);
336		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
337		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
338
339#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 
 
 
 
340		if (tw->tw_family == PF_INET6) {
341			struct ipv6_pinfo *np = inet6_sk(sk);
342			struct inet6_timewait_sock *tw6;
343
344			tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
345			tw6 = inet6_twsk((struct sock *)tw);
346			ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
347			ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
348			tw->tw_ipv6only = np->ipv6only;
349		}
350#endif
351
352#ifdef CONFIG_TCP_MD5SIG
353		/*
354		 * The timewait bucket does not have the key DB from the
355		 * sock structure. We just make a quick copy of the
356		 * md5 key being used (if indeed we are using one)
357		 * so the timewait ack generating code has the key.
358		 */
359		do {
360			struct tcp_md5sig_key *key;
361			memset(tcptw->tw_md5_key, 0, sizeof(tcptw->tw_md5_key));
362			tcptw->tw_md5_keylen = 0;
363			key = tp->af_specific->md5_lookup(sk, sk);
364			if (key != NULL) {
365				memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
366				tcptw->tw_md5_keylen = key->keylen;
367				if (tcp_alloc_md5sig_pool(sk) == NULL)
368					BUG();
369			}
370		} while (0);
371#endif
372
373		/* Linkage updates. */
374		__inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
375
376		/* Get the TIME_WAIT timeout firing. */
377		if (timeo < rto)
378			timeo = rto;
379
380		if (recycle_ok) {
381			tw->tw_timeout = rto;
382		} else {
383			tw->tw_timeout = TCP_TIMEWAIT_LEN;
384			if (state == TCP_TIME_WAIT)
385				timeo = TCP_TIMEWAIT_LEN;
386		}
387
388		inet_twsk_schedule(tw, &tcp_death_row, timeo,
389				   TCP_TIMEWAIT_LEN);
390		inet_twsk_put(tw);
 
 
 
 
 
 
 
 
391	} else {
392		/* Sorry, if we're out of memory, just CLOSE this
393		 * socket up.  We've got bigger problems than
394		 * non-graceful socket closings.
395		 */
396		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
397	}
398
399	tcp_update_metrics(sk);
400	tcp_done(sk);
401}
 
 
 
 
 
 
 
 
 
 
 
 
 
402
403void tcp_twsk_destructor(struct sock *sk)
404{
405#ifdef CONFIG_TCP_MD5SIG
406	struct tcp_timewait_sock *twsk = tcp_twsk(sk);
407	if (twsk->tw_md5_keylen)
408		tcp_free_md5sig_pool();
 
 
 
409#endif
 
410}
411EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
412
413static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
414					 struct request_sock *req)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
415{
416	tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
417}
418
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
419/* This is not only more efficient than what we used to do, it eliminates
420 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
421 *
422 * Actually, we could lots of memory writes here. tp of listening
423 * socket contains all necessary default parameters.
424 */
425struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
 
 
426{
427	struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
 
 
 
 
 
 
 
 
 
428
429	if (newsk != NULL) {
430		const struct inet_request_sock *ireq = inet_rsk(req);
431		struct tcp_request_sock *treq = tcp_rsk(req);
432		struct inet_connection_sock *newicsk = inet_csk(newsk);
433		struct tcp_sock *newtp = tcp_sk(newsk);
434		struct tcp_sock *oldtp = tcp_sk(sk);
435		struct tcp_cookie_values *oldcvp = oldtp->cookie_values;
436
437		/* TCP Cookie Transactions require space for the cookie pair,
438		 * as it differs for each connection.  There is no need to
439		 * copy any s_data_payload stored at the original socket.
440		 * Failure will prevent resuming the connection.
441		 *
442		 * Presumed copied, in order of appearance:
443		 *	cookie_in_always, cookie_out_never
444		 */
445		if (oldcvp != NULL) {
446			struct tcp_cookie_values *newcvp =
447				kzalloc(sizeof(*newtp->cookie_values),
448					GFP_ATOMIC);
449
450			if (newcvp != NULL) {
451				kref_init(&newcvp->kref);
452				newcvp->cookie_desired =
453						oldcvp->cookie_desired;
454				newtp->cookie_values = newcvp;
455			} else {
456				/* Not Yet Implemented */
457				newtp->cookie_values = NULL;
458			}
459		}
460
461		/* Now setup tcp_sock */
462		newtp->pred_flags = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
463
464		newtp->rcv_wup = newtp->copied_seq =
465		newtp->rcv_nxt = treq->rcv_isn + 1;
466
467		newtp->snd_sml = newtp->snd_una =
468		newtp->snd_nxt = newtp->snd_up =
469			treq->snt_isn + 1 + tcp_s_data_size(oldtp);
470
471		tcp_prequeue_init(newtp);
472
473		tcp_init_wl(newtp, treq->rcv_isn);
474
475		newtp->srtt = 0;
476		newtp->mdev = TCP_TIMEOUT_INIT;
477		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
478
479		newtp->packets_out = 0;
480		newtp->retrans_out = 0;
481		newtp->sacked_out = 0;
482		newtp->fackets_out = 0;
483		newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
484
485		/* So many TCP implementations out there (incorrectly) count the
486		 * initial SYN frame in their delayed-ACK and congestion control
487		 * algorithms that we must have the following bandaid to talk
488		 * efficiently to them.  -DaveM
489		 */
490		newtp->snd_cwnd = TCP_INIT_CWND;
491		newtp->snd_cwnd_cnt = 0;
492		newtp->bytes_acked = 0;
493
494		newtp->frto_counter = 0;
495		newtp->frto_highmark = 0;
496
497		newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
498
499		tcp_set_ca_state(newsk, TCP_CA_Open);
500		tcp_init_xmit_timers(newsk);
501		skb_queue_head_init(&newtp->out_of_order_queue);
502		newtp->write_seq = newtp->pushed_seq =
503			treq->snt_isn + 1 + tcp_s_data_size(oldtp);
504
505		newtp->rx_opt.saw_tstamp = 0;
506
507		newtp->rx_opt.dsack = 0;
508		newtp->rx_opt.num_sacks = 0;
509
510		newtp->urg_data = 0;
511
512		if (sock_flag(newsk, SOCK_KEEPOPEN))
513			inet_csk_reset_keepalive_timer(newsk,
514						       keepalive_time_when(newtp));
515
516		newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
517		if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
518			if (sysctl_tcp_fack)
519				tcp_enable_fack(newtp);
520		}
521		newtp->window_clamp = req->window_clamp;
522		newtp->rcv_ssthresh = req->rcv_wnd;
523		newtp->rcv_wnd = req->rcv_wnd;
524		newtp->rx_opt.wscale_ok = ireq->wscale_ok;
525		if (newtp->rx_opt.wscale_ok) {
526			newtp->rx_opt.snd_wscale = ireq->snd_wscale;
527			newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
528		} else {
529			newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
530			newtp->window_clamp = min(newtp->window_clamp, 65535U);
531		}
532		newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
533				  newtp->rx_opt.snd_wscale);
534		newtp->max_window = newtp->snd_wnd;
535
536		if (newtp->rx_opt.tstamp_ok) {
537			newtp->rx_opt.ts_recent = req->ts_recent;
538			newtp->rx_opt.ts_recent_stamp = get_seconds();
539			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
540		} else {
541			newtp->rx_opt.ts_recent_stamp = 0;
542			newtp->tcp_header_len = sizeof(struct tcphdr);
543		}
 
 
 
544#ifdef CONFIG_TCP_MD5SIG
545		newtp->md5sig_info = NULL;	/*XXX*/
546		if (newtp->af_specific->md5_lookup(sk, newsk))
547			newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
548#endif
549		if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
550			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
551		newtp->rx_opt.mss_clamp = req->mss;
552		TCP_ECN_openreq_child(newtp, req);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
553
554		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
555	}
556	return newsk;
557}
558EXPORT_SYMBOL(tcp_create_openreq_child);
559
560/*
561 *	Process an incoming packet for SYN_RECV sockets represented
562 *	as a request_sock.
 
 
 
 
 
 
 
 
 
563 */
564
565struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
566			   struct request_sock *req,
567			   struct request_sock **prev)
568{
569	struct tcp_options_received tmp_opt;
570	u8 *hash_location;
571	struct sock *child;
572	const struct tcphdr *th = tcp_hdr(skb);
573	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
574	int paws_reject = 0;
 
575
576	tmp_opt.saw_tstamp = 0;
577	if (th->doff > (sizeof(struct tcphdr)>>2)) {
578		tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
579
580		if (tmp_opt.saw_tstamp) {
581			tmp_opt.ts_recent = req->ts_recent;
 
 
582			/* We do not store true stamp, but it is not required,
583			 * it can be estimated (approximately)
584			 * from another data.
585			 */
586			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
587			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
588		}
589	}
590
591	/* Check for pure retransmitted SYN. */
592	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
593	    flg == TCP_FLAG_SYN &&
594	    !paws_reject) {
595		/*
596		 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
597		 * this case on figure 6 and figure 8, but formal
598		 * protocol description says NOTHING.
599		 * To be more exact, it says that we should send ACK,
600		 * because this segment (at least, if it has no data)
601		 * is out of window.
602		 *
603		 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
604		 *  describe SYN-RECV state. All the description
605		 *  is wrong, we cannot believe to it and should
606		 *  rely only on common sense and implementation
607		 *  experience.
608		 *
609		 * Enforce "SYN-ACK" according to figure 8, figure 6
610		 * of RFC793, fixed by RFC1122.
 
 
 
 
 
 
611		 */
612		req->rsk_ops->rtx_syn_ack(sk, req, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
613		return NULL;
614	}
615
616	/* Further reproduces section "SEGMENT ARRIVES"
617	   for state SYN-RECEIVED of RFC793.
618	   It is broken, however, it does not work only
619	   when SYNs are crossed.
620
621	   You would think that SYN crossing is impossible here, since
622	   we should have a SYN_SENT socket (from connect()) on our end,
623	   but this is not true if the crossed SYNs were sent to both
624	   ends by a malicious third party.  We must defend against this,
625	   and to do that we first verify the ACK (as per RFC793, page
626	   36) and reset if it is invalid.  Is this a true full defense?
627	   To convince ourselves, let us consider a way in which the ACK
628	   test can still pass in this 'malicious crossed SYNs' case.
629	   Malicious sender sends identical SYNs (and thus identical sequence
630	   numbers) to both A and B:
631
632		A: gets SYN, seq=7
633		B: gets SYN, seq=7
634
635	   By our good fortune, both A and B select the same initial
636	   send sequence number of seven :-)
637
638		A: sends SYN|ACK, seq=7, ack_seq=8
639		B: sends SYN|ACK, seq=7, ack_seq=8
640
641	   So we are now A eating this SYN|ACK, ACK test passes.  So
642	   does sequence test, SYN is truncated, and thus we consider
643	   it a bare ACK.
644
645	   If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
646	   bare ACK.  Otherwise, we create an established connection.  Both
647	   ends (listening sockets) accept the new incoming connection and try
648	   to talk to each other. 8-)
649
650	   Note: This case is both harmless, and rare.  Possibility is about the
651	   same as us discovering intelligent life on another plant tomorrow.
652
653	   But generally, we should (RFC lies!) to accept ACK
654	   from SYNACK both here and in tcp_rcv_state_process().
655	   tcp_rcv_state_process() does not, hence, we do not too.
656
657	   Note that the case is absolutely generic:
658	   we cannot optimize anything here without
659	   violating protocol. All the checks must be made
660	   before attempt to create socket.
661	 */
662
663	/* RFC793 page 36: "If the connection is in any non-synchronized state ...
664	 *                  and the incoming segment acknowledges something not yet
665	 *                  sent (the segment carries an unacceptable ACK) ...
666	 *                  a reset is sent."
667	 *
668	 * Invalid ACK: reset will be sent by listening socket
 
 
 
669	 */
670	if ((flg & TCP_FLAG_ACK) &&
671	    (TCP_SKB_CB(skb)->ack_seq !=
672	     tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk))))
673		return sk;
674
675	/* Also, it would be not so bad idea to check rcv_tsecr, which
676	 * is essentially ACK extension and too early or too late values
677	 * should cause reset in unsynchronized states.
678	 */
679
680	/* RFC793: "first check sequence number". */
681
682	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
683					  tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) {
684		/* Out of window: send ACK and drop. */
685		if (!(flg & TCP_FLAG_RST))
 
 
 
686			req->rsk_ops->send_ack(sk, skb, req);
687		if (paws_reject)
688			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
689		return NULL;
690	}
691
692	/* In sequence, PAWS is OK. */
693
694	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1))
695		req->ts_recent = tmp_opt.rcv_tsval;
 
 
 
696
697	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
698		/* Truncate SYN, it is out of window starting
699		   at tcp_rsk(req)->rcv_isn + 1. */
700		flg &= ~TCP_FLAG_SYN;
701	}
702
703	/* RFC793: "second check the RST bit" and
704	 *	   "fourth, check the SYN bit"
705	 */
706	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
707		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
708		goto embryonic_reset;
709	}
710
711	/* ACK sequence verified above, just make sure ACK is
712	 * set.  If ACK not set, just silently drop the packet.
 
 
 
713	 */
714	if (!(flg & TCP_FLAG_ACK))
715		return NULL;
716
 
 
 
 
 
 
717	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
718	if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
719	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
720		inet_rsk(req)->acked = 1;
721		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
722		return NULL;
723	}
724	if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
725		tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
726	else if (req->retrans) /* don't take RTT sample if retrans && ~TS */
727		tcp_rsk(req)->snt_synack = 0;
728
729	/* OK, ACK is valid, create big socket and
730	 * feed this segment to it. It will repeat all
731	 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
732	 * ESTABLISHED STATE. If it will be dropped after
733	 * socket is created, wait for troubles.
734	 */
735	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
736	if (child == NULL)
 
737		goto listen_overflow;
738
739	inet_csk_reqsk_queue_unlink(sk, req, prev);
740	inet_csk_reqsk_queue_removed(sk, req);
 
 
 
741
742	inet_csk_reqsk_queue_add(sk, req, child);
743	return child;
 
 
744
745listen_overflow:
746	if (!sysctl_tcp_abort_on_overflow) {
 
 
 
747		inet_rsk(req)->acked = 1;
748		return NULL;
749	}
750
751embryonic_reset:
752	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
753	if (!(flg & TCP_FLAG_RST))
 
 
 
 
754		req->rsk_ops->send_reset(sk, skb);
755
756	inet_csk_reqsk_queue_drop(sk, req, prev);
 
 
 
 
 
 
 
 
 
757	return NULL;
758}
759EXPORT_SYMBOL(tcp_check_req);
760
761/*
762 * Queue segment on the new socket if the new socket is active,
763 * otherwise we just shortcircuit this and continue with
764 * the new socket.
 
 
 
 
 
 
765 */
766
767int tcp_child_process(struct sock *parent, struct sock *child,
768		      struct sk_buff *skb)
 
769{
770	int ret = 0;
771	int state = child->sk_state;
772
 
 
 
 
773	if (!sock_owned_by_user(child)) {
774		ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
775					    skb->len);
776		/* Wakeup parent, send SIGIO */
777		if (state == TCP_SYN_RECV && child->sk_state != state)
778			parent->sk_data_ready(parent, 0);
779	} else {
780		/* Alas, it is possible again, because we do lookup
781		 * in main socket hash table and lock on listening
782		 * socket does not protect us more.
783		 */
784		__sk_add_backlog(child, skb);
785	}
786
787	bh_unlock_sock(child);
788	sock_put(child);
789	return ret;
790}
791EXPORT_SYMBOL(tcp_child_process);