1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: stable/10/sys/netinet/tcp_timewait.c 309108 2016-11-24 14:48:46Z jch $");
34
35#include "opt_inet.h"
36#include "opt_inet6.h"
37#include "opt_tcpdebug.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/callout.h>
42#include <sys/kernel.h>
43#include <sys/sysctl.h>
44#include <sys/malloc.h>
45#include <sys/mbuf.h>
46#include <sys/priv.h>
47#include <sys/proc.h>
48#include <sys/socket.h>
49#include <sys/socketvar.h>
50#include <sys/protosw.h>
51#include <sys/random.h>
52
53#include <vm/uma.h>
54
55#include <net/route.h>
56#include <net/if.h>
57#include <net/vnet.h>
58
59#include <netinet/in.h>
60#include <netinet/in_pcb.h>
61#include <netinet/in_systm.h>
62#include <netinet/in_var.h>
63#include <netinet/ip.h>
64#include <netinet/ip_icmp.h>
65#include <netinet/ip_var.h>
66#ifdef INET6
67#include <netinet/ip6.h>
68#include <netinet6/in6_pcb.h>
69#include <netinet6/ip6_var.h>
70#include <netinet6/scope6_var.h>
71#include <netinet6/nd6.h>
72#endif
73#include <netinet/tcp.h>
74#include <netinet/tcp_fsm.h>
75#include <netinet/tcp_seq.h>
76#include <netinet/tcp_timer.h>
77#include <netinet/tcp_var.h>
78#ifdef INET6
79#include <netinet6/tcp6_var.h>
80#endif
81#include <netinet/tcpip.h>
82#ifdef TCPDEBUG
83#include <netinet/tcp_debug.h>
84#endif
85#ifdef INET6
86#include <netinet6/ip6protosw.h>
87#endif
88
89#include <machine/in_cksum.h>
90
91#include <security/mac/mac_framework.h>
92
93static VNET_DEFINE(uma_zone_t, tcptw_zone);
94#define	V_tcptw_zone		VNET(tcptw_zone)
95static int	maxtcptw;
96
97/*
98 * The timed wait queue contains references to each of the TCP sessions
99 * currently in the TIME_WAIT state.  The queue pointers, including the
100 * queue pointers in each tcptw structure, are protected using the global
101 * timewait lock, which must be held over queue iteration and modification.
102 *
103 * Rules on tcptw usage:
104 *  - a inpcb is always freed _after_ its tcptw
105 *  - a tcptw relies on its inpcb reference counting for memory stability
106 *  - a tcptw is dereferenceable only while its inpcb is locked
107 */
108static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
109#define	V_twq_2msl		VNET(twq_2msl)
110
111/* Global timewait lock */
112static VNET_DEFINE(struct rwlock, tw_lock);
113#define	V_tw_lock		VNET(tw_lock)
114
115#define	TW_LOCK_INIT(tw, d)	rw_init_flags(&(tw), (d), 0)
116#define	TW_LOCK_DESTROY(tw)	rw_destroy(&(tw))
117#define	TW_RLOCK(tw)		rw_rlock(&(tw))
118#define	TW_WLOCK(tw)		rw_wlock(&(tw))
119#define	TW_RUNLOCK(tw)		rw_runlock(&(tw))
120#define	TW_WUNLOCK(tw)		rw_wunlock(&(tw))
121#define	TW_LOCK_ASSERT(tw)	rw_assert(&(tw), RA_LOCKED)
122#define	TW_RLOCK_ASSERT(tw)	rw_assert(&(tw), RA_RLOCKED)
123#define	TW_WLOCK_ASSERT(tw)	rw_assert(&(tw), RA_WLOCKED)
124#define	TW_UNLOCK_ASSERT(tw)	rw_assert(&(tw), RA_UNLOCKED)
125
126static void	tcp_tw_2msl_reset(struct tcptw *, int);
127static void	tcp_tw_2msl_stop(struct tcptw *, int);
128static int	tcp_twrespond(struct tcptw *, int);
129
130static int
131tcptw_auto_size(void)
132{
133	int halfrange;
134
135	/*
136	 * Max out at half the ephemeral port range so that TIME_WAIT
137	 * sockets don't tie up too many ephemeral ports.
138	 */
139	if (V_ipport_lastauto > V_ipport_firstauto)
140		halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
141	else
142		halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
143	/* Protect against goofy port ranges smaller than 32. */
144	return (imin(imax(halfrange, 32), maxsockets / 5));
145}
146
147static int
148sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
149{
150	int error, new;
151
152	if (maxtcptw == 0)
153		new = tcptw_auto_size();
154	else
155		new = maxtcptw;
156	error = sysctl_handle_int(oidp, &new, 0, req);
157	if (error == 0 && req->newptr)
158		if (new >= 32) {
159			maxtcptw = new;
160			uma_zone_set_max(V_tcptw_zone, maxtcptw);
161		}
162	return (error);
163}
164
165SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
166    &maxtcptw, 0, sysctl_maxtcptw, "IU",
167    "Maximum number of compressed TCP TIME_WAIT entries");
168
169VNET_DEFINE(int, nolocaltimewait) = 0;
170#define	V_nolocaltimewait	VNET(nolocaltimewait)
171SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
172    &VNET_NAME(nolocaltimewait), 0,
173    "Do not create compressed TCP TIME_WAIT entries for local connections");
174
175void
176tcp_tw_zone_change(void)
177{
178
179	if (maxtcptw == 0)
180		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
181}
182
183void
184tcp_tw_init(void)
185{
186
187	V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
188	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
189	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
190	if (maxtcptw == 0)
191		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
192	else
193		uma_zone_set_max(V_tcptw_zone, maxtcptw);
194	TAILQ_INIT(&V_twq_2msl);
195	TW_LOCK_INIT(V_tw_lock, "tcptw");
196}
197
198#ifdef VIMAGE
199void
200tcp_tw_destroy(void)
201{
202	struct tcptw *tw;
203
204	INP_INFO_RLOCK(&V_tcbinfo);
205	while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
206		tcp_twclose(tw, 0);
207	INP_INFO_RUNLOCK(&V_tcbinfo);
208
209	TW_LOCK_DESTROY(V_tw_lock);
210	uma_zdestroy(V_tcptw_zone);
211}
212#endif
213
214/*
215 * Move a TCP connection into TIME_WAIT state.
216 *    tcbinfo is locked.
217 *    inp is locked, and is unlocked before returning.
218 */
219void
220tcp_twstart(struct tcpcb *tp)
221{
222	struct tcptw *tw;
223	struct inpcb *inp = tp->t_inpcb;
224	int acknow;
225	struct socket *so;
226#ifdef INET6
227	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
228#endif
229
230	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
231	INP_WLOCK_ASSERT(inp);
232
233	/* A dropped inp should never transition to TIME_WAIT state. */
234	KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
235	    "(inp->inp_flags & INP_DROPPED) != 0"));
236
237	if (V_nolocaltimewait) {
238		int error = 0;
239#ifdef INET6
240		if (isipv6)
241			error = in6_localaddr(&inp->in6p_faddr);
242#endif
243#if defined(INET6) && defined(INET)
244		else
245#endif
246#ifdef INET
247			error = in_localip(inp->inp_faddr);
248#endif
249		if (error) {
250			tp = tcp_close(tp);
251			if (tp != NULL)
252				INP_WUNLOCK(inp);
253			return;
254		}
255	}
256
257
258	/*
259	 * For use only by DTrace.  We do not reference the state
260	 * after this point so modifying it in place is not a problem.
261	 */
262	tcp_state_change(tp, TCPS_TIME_WAIT);
263
264	tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
265	if (tw == NULL) {
266		/*
267		 * Reached limit on total number of TIMEWAIT connections
268		 * allowed. Remove a connection from TIMEWAIT queue in LRU
269		 * fashion to make room for this connection.
270		 *
271		 * XXX:  Check if it possible to always have enough room
272		 * in advance based on guarantees provided by uma_zalloc().
273		 */
274		tw = tcp_tw_2msl_scan(1);
275		if (tw == NULL) {
276			tp = tcp_close(tp);
277			if (tp != NULL)
278				INP_WUNLOCK(inp);
279			return;
280		}
281	}
282	/*
283	 * The tcptw will hold a reference on its inpcb until tcp_twclose
284	 * is called
285	 */
286	tw->tw_inpcb = inp;
287	in_pcbref(inp);	/* Reference from tw */
288
289	/*
290	 * Recover last window size sent.
291	 */
292	if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
293		tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
294	else
295		tw->last_win = 0;
296
297	/*
298	 * Set t_recent if timestamps are used on the connection.
299	 */
300	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
301	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
302		tw->t_recent = tp->ts_recent;
303		tw->ts_offset = tp->ts_offset;
304	} else {
305		tw->t_recent = 0;
306		tw->ts_offset = 0;
307	}
308
309	tw->snd_nxt = tp->snd_nxt;
310	tw->rcv_nxt = tp->rcv_nxt;
311	tw->iss     = tp->iss;
312	tw->irs     = tp->irs;
313	tw->t_starttime = tp->t_starttime;
314	tw->tw_time = 0;
315
316/* XXX
317 * If this code will
318 * be used for fin-wait-2 state also, then we may need
319 * a ts_recent from the last segment.
320 */
321	acknow = tp->t_flags & TF_ACKNOW;
322
323	/*
324	 * First, discard tcpcb state, which includes stopping its timers and
325	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
326	 * that work is now done in the caller.
327	 *
328	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
329	 * and might not be needed here any longer.
330	 */
331	tcp_discardcb(tp);
332	so = inp->inp_socket;
333	soisdisconnected(so);
334	tw->tw_cred = crhold(so->so_cred);
335	SOCK_LOCK(so);
336	tw->tw_so_options = so->so_options;
337	SOCK_UNLOCK(so);
338	if (acknow)
339		tcp_twrespond(tw, TH_ACK);
340	inp->inp_ppcb = tw;
341	inp->inp_flags |= INP_TIMEWAIT;
342	tcp_tw_2msl_reset(tw, 0);
343
344	/*
345	 * If the inpcb owns the sole reference to the socket, then we can
346	 * detach and free the socket as it is not needed in time wait.
347	 */
348	if (inp->inp_flags & INP_SOCKREF) {
349		KASSERT(so->so_state & SS_PROTOREF,
350		    ("tcp_twstart: !SS_PROTOREF"));
351		inp->inp_flags &= ~INP_SOCKREF;
352		INP_WUNLOCK(inp);
353		ACCEPT_LOCK();
354		SOCK_LOCK(so);
355		so->so_state &= ~SS_PROTOREF;
356		sofree(so);
357	} else
358		INP_WUNLOCK(inp);
359}
360
361#if 0
362/*
363 * The appromixate rate of ISN increase of Microsoft TCP stacks;
364 * the actual rate is slightly higher due to the addition of
365 * random positive increments.
366 *
367 * Most other new OSes use semi-randomized ISN values, so we
368 * do not need to worry about them.
369 */
370#define	MS_ISN_BYTES_PER_SECOND		250000
371
372/*
373 * Determine if the ISN we will generate has advanced beyond the last
374 * sequence number used by the previous connection.  If so, indicate
375 * that it is safe to recycle this tw socket by returning 1.
376 */
377int
378tcp_twrecycleable(struct tcptw *tw)
379{
380	tcp_seq new_iss = tw->iss;
381	tcp_seq new_irs = tw->irs;
382
383	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
384	new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
385	new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
386
387	if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
388		return (1);
389	else
390		return (0);
391}
392#endif
393
394/*
395 * Returns 1 if the TIME_WAIT state was killed and we should start over,
396 * looking for a pcb in the listen state.  Returns 0 otherwise.
397 */
398int
399tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
400    struct mbuf *m, int tlen)
401{
402	struct tcptw *tw;
403	int thflags;
404	tcp_seq seq;
405
406	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
407	INP_WLOCK_ASSERT(inp);
408
409	/*
410	 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
411	 * still present.  This is undesirable, but temporarily necessary
412	 * until we work out how to handle inpcb's who's timewait state has
413	 * been removed.
414	 */
415	tw = intotw(inp);
416	if (tw == NULL)
417		goto drop;
418
419	thflags = th->th_flags;
420
421	/*
422	 * NOTE: for FIN_WAIT_2 (to be added later),
423	 * must validate sequence number before accepting RST
424	 */
425
426	/*
427	 * If the segment contains RST:
428	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
429	 *      RFC 1337.
430	 */
431	if (thflags & TH_RST)
432		goto drop;
433
434#if 0
435/* PAWS not needed at the moment */
436	/*
437	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
438	 * and it's less than ts_recent, drop it.
439	 */
440	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
441	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
442		if ((thflags & TH_ACK) == 0)
443			goto drop;
444		goto ack;
445	}
446	/*
447	 * ts_recent is never updated because we never accept new segments.
448	 */
449#endif
450
451	/*
452	 * If a new connection request is received
453	 * while in TIME_WAIT, drop the old connection
454	 * and start over if the sequence numbers
455	 * are above the previous ones.
456	 */
457	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
458		tcp_twclose(tw, 0);
459		return (1);
460	}
461
462	/*
463	 * Drop the segment if it does not contain an ACK.
464	 */
465	if ((thflags & TH_ACK) == 0)
466		goto drop;
467
468	/*
469	 * Reset the 2MSL timer if this is a duplicate FIN.
470	 */
471	if (thflags & TH_FIN) {
472		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
473		if (seq + 1 == tw->rcv_nxt)
474			tcp_tw_2msl_reset(tw, 1);
475	}
476
477	/*
478	 * Acknowledge the segment if it has data or is not a duplicate ACK.
479	 */
480	if (thflags != TH_ACK || tlen != 0 ||
481	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
482		tcp_twrespond(tw, TH_ACK);
483drop:
484	INP_WUNLOCK(inp);
485	m_freem(m);
486	return (0);
487}
488
489void
490tcp_twclose(struct tcptw *tw, int reuse)
491{
492	struct socket *so;
493	struct inpcb *inp;
494
495	/*
496	 * At this point, we are in one of two situations:
497	 *
498	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
499	 *     all state.
500	 *
501	 * (2) We have a socket -- if we own a reference, release it and
502	 *     notify the socket layer.
503	 */
504	inp = tw->tw_inpcb;
505	KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
506	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
507	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);	/* in_pcbfree() */
508	INP_WLOCK_ASSERT(inp);
509
510	tcp_tw_2msl_stop(tw, reuse);
511	inp->inp_ppcb = NULL;
512	in_pcbdrop(inp);
513
514	so = inp->inp_socket;
515	if (so != NULL) {
516		/*
517		 * If there's a socket, handle two cases: first, we own a
518		 * strong reference, which we will now release, or we don't
519		 * in which case another reference exists (XXXRW: think
520		 * about this more), and we don't need to take action.
521		 */
522		if (inp->inp_flags & INP_SOCKREF) {
523			inp->inp_flags &= ~INP_SOCKREF;
524			INP_WUNLOCK(inp);
525			ACCEPT_LOCK();
526			SOCK_LOCK(so);
527			KASSERT(so->so_state & SS_PROTOREF,
528			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
529			so->so_state &= ~SS_PROTOREF;
530			sofree(so);
531		} else {
532			/*
533			 * If we don't own the only reference, the socket and
534			 * inpcb need to be left around to be handled by
535			 * tcp_usr_detach() later.
536			 */
537			INP_WUNLOCK(inp);
538		}
539	} else {
540		/*
541		 * The socket has been already cleaned-up for us, only free the
542		 * inpcb.
543		 */
544		in_pcbfree(inp);
545	}
546	TCPSTAT_INC(tcps_closed);
547}
548
549static int
550tcp_twrespond(struct tcptw *tw, int flags)
551{
552	struct inpcb *inp = tw->tw_inpcb;
553#if defined(INET6) || defined(INET)
554	struct tcphdr *th = NULL;
555#endif
556	struct mbuf *m;
557#ifdef INET
558	struct ip *ip = NULL;
559#endif
560	u_int hdrlen, optlen;
561	int error = 0;			/* Keep compiler happy */
562	struct tcpopt to;
563#ifdef INET6
564	struct ip6_hdr *ip6 = NULL;
565	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
566#endif
567	hdrlen = 0;                     /* Keep compiler happy */
568
569	INP_WLOCK_ASSERT(inp);
570
571	m = m_gethdr(M_NOWAIT, MT_DATA);
572	if (m == NULL)
573		return (ENOBUFS);
574	m->m_data += max_linkhdr;
575
576#ifdef MAC
577	mac_inpcb_create_mbuf(inp, m);
578#endif
579
580#ifdef INET6
581	if (isipv6) {
582		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
583		ip6 = mtod(m, struct ip6_hdr *);
584		th = (struct tcphdr *)(ip6 + 1);
585		tcpip_fillheaders(inp, ip6, th);
586	}
587#endif
588#if defined(INET6) && defined(INET)
589	else
590#endif
591#ifdef INET
592	{
593		hdrlen = sizeof(struct tcpiphdr);
594		ip = mtod(m, struct ip *);
595		th = (struct tcphdr *)(ip + 1);
596		tcpip_fillheaders(inp, ip, th);
597	}
598#endif
599	to.to_flags = 0;
600
601	/*
602	 * Send a timestamp and echo-reply if both our side and our peer
603	 * have sent timestamps in our SYN's and this is not a RST.
604	 */
605	if (tw->t_recent && flags == TH_ACK) {
606		to.to_flags |= TOF_TS;
607		to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
608		to.to_tsecr = tw->t_recent;
609	}
610	optlen = tcp_addoptions(&to, (u_char *)(th + 1));
611
612	m->m_len = hdrlen + optlen;
613	m->m_pkthdr.len = m->m_len;
614
615	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
616
617	th->th_seq = htonl(tw->snd_nxt);
618	th->th_ack = htonl(tw->rcv_nxt);
619	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
620	th->th_flags = flags;
621	th->th_win = htons(tw->last_win);
622
623	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
624#ifdef INET6
625	if (isipv6) {
626		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
627		th->th_sum = in6_cksum_pseudo(ip6,
628		    sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
629		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
630		error = ip6_output(m, inp->in6p_outputopts, NULL,
631		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
632	}
633#endif
634#if defined(INET6) && defined(INET)
635	else
636#endif
637#ifdef INET
638	{
639		m->m_pkthdr.csum_flags = CSUM_TCP;
640		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
641		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
642		ip->ip_len = htons(m->m_pkthdr.len);
643		if (V_path_mtu_discovery)
644			ip->ip_off |= htons(IP_DF);
645		error = ip_output(m, inp->inp_options, NULL,
646		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
647		    NULL, inp);
648	}
649#endif
650	if (flags & TH_ACK)
651		TCPSTAT_INC(tcps_sndacks);
652	else
653		TCPSTAT_INC(tcps_sndctrl);
654	TCPSTAT_INC(tcps_sndtotal);
655	return (error);
656}
657
658static void
659tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
660{
661
662	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
663	INP_WLOCK_ASSERT(tw->tw_inpcb);
664
665	TW_WLOCK(V_tw_lock);
666	if (rearm)
667		TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
668	tw->tw_time = ticks + 2 * tcp_msl;
669	TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
670	TW_WUNLOCK(V_tw_lock);
671}
672
673static void
674tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
675{
676	struct ucred *cred;
677	struct inpcb *inp;
678	int released;
679
680	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
681
682	TW_WLOCK(V_tw_lock);
683	inp = tw->tw_inpcb;
684	tw->tw_inpcb = NULL;
685
686	TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
687	cred = tw->tw_cred;
688	tw->tw_cred = NULL;
689	TW_WUNLOCK(V_tw_lock);
690
691	if (cred != NULL)
692		crfree(cred);
693
694	released = in_pcbrele_wlocked(inp);
695	KASSERT(!released, ("%s: inp should not be released here", __func__));
696
697	if (!reuse)
698		uma_zfree(V_tcptw_zone, tw);
699}
700
701struct tcptw *
702tcp_tw_2msl_scan(int reuse)
703{
704	struct tcptw *tw;
705	struct inpcb *inp;
706
707#ifdef INVARIANTS
708	if (reuse) {
709		/*
710		 * Exclusive pcbinfo lock is not required in reuse case even if
711		 * two inpcb locks can be acquired simultaneously:
712		 *  - the inpcb transitioning to TIME_WAIT state in
713		 *    tcp_tw_start(),
714		 *  - the inpcb closed by tcp_twclose().
715		 *
716		 * It is because only inpcbs in FIN_WAIT2 or CLOSING states can
717		 * transition in TIME_WAIT state.  Then a pcbcb cannot be in
718		 * TIME_WAIT list and transitioning to TIME_WAIT state at same
719		 * time.
720		 */
721		INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
722	}
723#endif
724
725	for (;;) {
726		TW_RLOCK(V_tw_lock);
727		tw = TAILQ_FIRST(&V_twq_2msl);
728		if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
729			TW_RUNLOCK(V_tw_lock);
730			break;
731		}
732		KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
733		    __func__));
734
735		inp = tw->tw_inpcb;
736		in_pcbref(inp);
737		TW_RUNLOCK(V_tw_lock);
738
739		if (INP_INFO_TRY_RLOCK(&V_tcbinfo)) {
740
741			INP_WLOCK(inp);
742			tw = intotw(inp);
743			if (in_pcbrele_wlocked(inp)) {
744				KASSERT(tw == NULL, ("%s: held last inp "
745				    "reference but tw not NULL", __func__));
746				INP_INFO_RUNLOCK(&V_tcbinfo);
747				continue;
748			}
749
750			if (tw == NULL) {
751				/* tcp_twclose() has already been called */
752				INP_WUNLOCK(inp);
753				INP_INFO_RUNLOCK(&V_tcbinfo);
754				continue;
755			}
756
757			tcp_twclose(tw, reuse);
758			INP_INFO_RUNLOCK(&V_tcbinfo);
759			if (reuse)
760			    return tw;
761		} else {
762			/* INP_INFO lock is busy, continue later. */
763			INP_WLOCK(inp);
764			if (!in_pcbrele_wlocked(inp))
765				INP_WUNLOCK(inp);
766			break;
767		}
768	}
769
770	return NULL;
771}
772