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/11/sys/netinet/tcp_timewait.c 368183 2020-11-30 10:58:06Z tuexen $");
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#ifndef INVARIANTS
51#include <sys/syslog.h>
52#endif
53#include <sys/protosw.h>
54#include <sys/random.h>
55
56#include <vm/uma.h>
57
58#include <net/route.h>
59#include <net/if.h>
60#include <net/if_var.h>
61#include <net/vnet.h>
62
63#include <netinet/in.h>
64#include <netinet/in_kdtrace.h>
65#include <netinet/in_pcb.h>
66#include <netinet/in_systm.h>
67#include <netinet/in_var.h>
68#include <netinet/ip.h>
69#include <netinet/ip_icmp.h>
70#include <netinet/ip_var.h>
71#ifdef INET6
72#include <netinet/ip6.h>
73#include <netinet6/in6_pcb.h>
74#include <netinet6/ip6_var.h>
75#include <netinet6/scope6_var.h>
76#include <netinet6/nd6.h>
77#endif
78#include <netinet/tcp.h>
79#include <netinet/tcp_fsm.h>
80#include <netinet/tcp_seq.h>
81#include <netinet/tcp_timer.h>
82#include <netinet/tcp_var.h>
83#ifdef INET6
84#include <netinet6/tcp6_var.h>
85#endif
86#include <netinet/tcpip.h>
87#ifdef TCPDEBUG
88#include <netinet/tcp_debug.h>
89#endif
90#ifdef INET6
91#include <netinet6/ip6protosw.h>
92#endif
93
94#include <machine/in_cksum.h>
95
96#include <security/mac/mac_framework.h>
97
98static VNET_DEFINE(uma_zone_t, tcptw_zone);
99#define	V_tcptw_zone		VNET(tcptw_zone)
100static int	maxtcptw;
101
102/*
103 * The timed wait queue contains references to each of the TCP sessions
104 * currently in the TIME_WAIT state.  The queue pointers, including the
105 * queue pointers in each tcptw structure, are protected using the global
106 * timewait lock, which must be held over queue iteration and modification.
107 *
108 * Rules on tcptw usage:
109 *  - a inpcb is always freed _after_ its tcptw
110 *  - a tcptw relies on its inpcb reference counting for memory stability
111 *  - a tcptw is dereferenceable only while its inpcb is locked
112 */
113static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
114#define	V_twq_2msl		VNET(twq_2msl)
115
116/* Global timewait lock */
117static VNET_DEFINE(struct rwlock, tw_lock);
118#define	V_tw_lock		VNET(tw_lock)
119
120#define	TW_LOCK_INIT(tw, d)	rw_init_flags(&(tw), (d), 0)
121#define	TW_LOCK_DESTROY(tw)	rw_destroy(&(tw))
122#define	TW_RLOCK(tw)		rw_rlock(&(tw))
123#define	TW_WLOCK(tw)		rw_wlock(&(tw))
124#define	TW_RUNLOCK(tw)		rw_runlock(&(tw))
125#define	TW_WUNLOCK(tw)		rw_wunlock(&(tw))
126#define	TW_LOCK_ASSERT(tw)	rw_assert(&(tw), RA_LOCKED)
127#define	TW_RLOCK_ASSERT(tw)	rw_assert(&(tw), RA_RLOCKED)
128#define	TW_WLOCK_ASSERT(tw)	rw_assert(&(tw), RA_WLOCKED)
129#define	TW_UNLOCK_ASSERT(tw)	rw_assert(&(tw), RA_UNLOCKED)
130
131static void	tcp_tw_2msl_reset(struct tcptw *, int);
132static void	tcp_tw_2msl_stop(struct tcptw *, int);
133static int	tcp_twrespond(struct tcptw *, int);
134
135static int
136tcptw_auto_size(void)
137{
138	int halfrange;
139
140	/*
141	 * Max out at half the ephemeral port range so that TIME_WAIT
142	 * sockets don't tie up too many ephemeral ports.
143	 */
144	if (V_ipport_lastauto > V_ipport_firstauto)
145		halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
146	else
147		halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
148	/* Protect against goofy port ranges smaller than 32. */
149	return (imin(imax(halfrange, 32), maxsockets / 5));
150}
151
152static int
153sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
154{
155	int error, new;
156
157	if (maxtcptw == 0)
158		new = tcptw_auto_size();
159	else
160		new = maxtcptw;
161	error = sysctl_handle_int(oidp, &new, 0, req);
162	if (error == 0 && req->newptr)
163		if (new >= 32) {
164			maxtcptw = new;
165			uma_zone_set_max(V_tcptw_zone, maxtcptw);
166		}
167	return (error);
168}
169
170SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
171    &maxtcptw, 0, sysctl_maxtcptw, "IU",
172    "Maximum number of compressed TCP TIME_WAIT entries");
173
174VNET_DEFINE(int, nolocaltimewait) = 0;
175#define	V_nolocaltimewait	VNET(nolocaltimewait)
176SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW,
177    &VNET_NAME(nolocaltimewait), 0,
178    "Do not create compressed TCP TIME_WAIT entries for local connections");
179
180void
181tcp_tw_zone_change(void)
182{
183
184	if (maxtcptw == 0)
185		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
186}
187
188void
189tcp_tw_init(void)
190{
191
192	V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
193	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
194	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
195	if (maxtcptw == 0)
196		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
197	else
198		uma_zone_set_max(V_tcptw_zone, maxtcptw);
199	TAILQ_INIT(&V_twq_2msl);
200	TW_LOCK_INIT(V_tw_lock, "tcptw");
201}
202
203#ifdef VIMAGE
204void
205tcp_tw_destroy(void)
206{
207	struct tcptw *tw;
208
209	INP_INFO_RLOCK(&V_tcbinfo);
210	while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
211		tcp_twclose(tw, 0);
212	INP_INFO_RUNLOCK(&V_tcbinfo);
213
214	TW_LOCK_DESTROY(V_tw_lock);
215	uma_zdestroy(V_tcptw_zone);
216}
217#endif
218
219/*
220 * Move a TCP connection into TIME_WAIT state.
221 *    tcbinfo is locked.
222 *    inp is locked, and is unlocked before returning.
223 */
224void
225tcp_twstart(struct tcpcb *tp)
226{
227	struct tcptw *tw;
228	struct inpcb *inp = tp->t_inpcb;
229	int acknow;
230	struct socket *so;
231	uint32_t recwin;
232#ifdef INET6
233	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
234#endif
235
236	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
237	INP_WLOCK_ASSERT(inp);
238
239	/* A dropped inp should never transition to TIME_WAIT state. */
240	KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
241	    "(inp->inp_flags & INP_DROPPED) != 0"));
242
243	if (V_nolocaltimewait) {
244		int error = 0;
245#ifdef INET6
246		if (isipv6)
247			error = in6_localaddr(&inp->in6p_faddr);
248#endif
249#if defined(INET6) && defined(INET)
250		else
251#endif
252#ifdef INET
253			error = in_localip(inp->inp_faddr);
254#endif
255		if (error) {
256			tp = tcp_close(tp);
257			if (tp != NULL)
258				INP_WUNLOCK(inp);
259			return;
260		}
261	}
262
263
264	/*
265	 * For use only by DTrace.  We do not reference the state
266	 * after this point so modifying it in place is not a problem.
267	 */
268	tcp_state_change(tp, TCPS_TIME_WAIT);
269
270	tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
271	if (tw == NULL) {
272		/*
273		 * Reached limit on total number of TIMEWAIT connections
274		 * allowed. Remove a connection from TIMEWAIT queue in LRU
275		 * fashion to make room for this connection.
276		 *
277		 * XXX:  Check if it possible to always have enough room
278		 * in advance based on guarantees provided by uma_zalloc().
279		 */
280		tw = tcp_tw_2msl_scan(1);
281		if (tw == NULL) {
282			tp = tcp_close(tp);
283			if (tp != NULL)
284				INP_WUNLOCK(inp);
285			return;
286		}
287	}
288	/*
289	 * The tcptw will hold a reference on its inpcb until tcp_twclose
290	 * is called
291	 */
292	tw->tw_inpcb = inp;
293	in_pcbref(inp);	/* Reference from tw */
294
295	/*
296	 * Recover last window size sent.
297	 */
298	so = inp->inp_socket;
299	recwin = lmin(lmax(sbspace(&so->so_rcv), 0),
300	    (long)TCP_MAXWIN << tp->rcv_scale);
301	if (recwin < (so->so_rcv.sb_hiwat / 4) &&
302	    recwin < tp->t_maxseg)
303		recwin = 0;
304	if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
305	    recwin < (tp->rcv_adv - tp->rcv_nxt))
306		recwin = (tp->rcv_adv - tp->rcv_nxt);
307	tw->last_win = (u_short)(recwin >> tp->rcv_scale);
308
309	/*
310	 * Set t_recent if timestamps are used on the connection.
311	 */
312	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
313	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
314		tw->t_recent = tp->ts_recent;
315		tw->ts_offset = tp->ts_offset;
316	} else {
317		tw->t_recent = 0;
318		tw->ts_offset = 0;
319	}
320
321	tw->snd_nxt = tp->snd_nxt;
322	tw->rcv_nxt = tp->rcv_nxt;
323	tw->iss     = tp->iss;
324	tw->irs     = tp->irs;
325	tw->t_starttime = tp->t_starttime;
326	tw->tw_time = 0;
327
328/* XXX
329 * If this code will
330 * be used for fin-wait-2 state also, then we may need
331 * a ts_recent from the last segment.
332 */
333	acknow = tp->t_flags & TF_ACKNOW;
334
335	/*
336	 * First, discard tcpcb state, which includes stopping its timers and
337	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
338	 * that work is now done in the caller.
339	 *
340	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
341	 * and might not be needed here any longer.
342	 */
343	tcp_discardcb(tp);
344	soisdisconnected(so);
345	tw->tw_cred = crhold(so->so_cred);
346	SOCK_LOCK(so);
347	tw->tw_so_options = so->so_options;
348	SOCK_UNLOCK(so);
349	if (acknow)
350		tcp_twrespond(tw, TH_ACK);
351	inp->inp_ppcb = tw;
352	inp->inp_flags |= INP_TIMEWAIT;
353	TCPSTATES_INC(TCPS_TIME_WAIT);
354	tcp_tw_2msl_reset(tw, 0);
355
356	/*
357	 * If the inpcb owns the sole reference to the socket, then we can
358	 * detach and free the socket as it is not needed in time wait.
359	 */
360	if (inp->inp_flags & INP_SOCKREF) {
361		KASSERT(so->so_state & SS_PROTOREF,
362		    ("tcp_twstart: !SS_PROTOREF"));
363		inp->inp_flags &= ~INP_SOCKREF;
364		INP_WUNLOCK(inp);
365		ACCEPT_LOCK();
366		SOCK_LOCK(so);
367		so->so_state &= ~SS_PROTOREF;
368		sofree(so);
369	} else
370		INP_WUNLOCK(inp);
371}
372
373/*
374 * Returns 1 if the TIME_WAIT state was killed and we should start over,
375 * looking for a pcb in the listen state.  Returns 0 otherwise.
376 * It be called with to == NULL only for pure SYN-segments.
377 */
378int
379tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
380    struct mbuf *m, int tlen)
381{
382	struct tcptw *tw;
383	int thflags;
384	tcp_seq seq;
385
386	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
387	INP_WLOCK_ASSERT(inp);
388
389	/*
390	 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
391	 * still present.  This is undesirable, but temporarily necessary
392	 * until we work out how to handle inpcb's who's timewait state has
393	 * been removed.
394	 */
395	tw = intotw(inp);
396	if (tw == NULL)
397		goto drop;
398
399	thflags = th->th_flags;
400	KASSERT(to != NULL || (thflags & (TH_SYN | TH_ACK)) == TH_SYN,
401	        ("tcp_twcheck: called without options on a non-SYN segment"));
402
403	/*
404	 * NOTE: for FIN_WAIT_2 (to be added later),
405	 * must validate sequence number before accepting RST
406	 */
407
408	/*
409	 * If the segment contains RST:
410	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
411	 *      RFC 1337.
412	 */
413	if (thflags & TH_RST)
414		goto drop;
415
416#if 0
417/* PAWS not needed at the moment */
418	/*
419	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
420	 * and it's less than ts_recent, drop it.
421	 */
422	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
423	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
424		if ((thflags & TH_ACK) == 0)
425			goto drop;
426		goto ack;
427	}
428	/*
429	 * ts_recent is never updated because we never accept new segments.
430	 */
431#endif
432
433	/*
434	 * If a new connection request is received
435	 * while in TIME_WAIT, drop the old connection
436	 * and start over if the sequence numbers
437	 * are above the previous ones.
438	 */
439	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
440		tcp_twclose(tw, 0);
441		return (1);
442	}
443
444	/*
445	 * Drop the segment if it does not contain an ACK.
446	 */
447	if ((thflags & TH_ACK) == 0)
448		goto drop;
449
450	/*
451	 * If timestamps were negotiated during SYN/ACK and a
452	 * segment without a timestamp is received, silently drop
453	 * the segment.
454	 * See section 3.2 of RFC 7323.
455	 */
456	if (((to->to_flags & TOF_TS) == 0) && (tw->t_recent != 0)) {
457		goto drop;
458	}
459
460	/*
461	 * Reset the 2MSL timer if this is a duplicate FIN.
462	 */
463	if (thflags & TH_FIN) {
464		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
465		if (seq + 1 == tw->rcv_nxt)
466			tcp_tw_2msl_reset(tw, 1);
467	}
468
469	/*
470	 * Acknowledge the segment if it has data or is not a duplicate ACK.
471	 */
472	if (thflags != TH_ACK || tlen != 0 ||
473	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) {
474		TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
475		tcp_twrespond(tw, TH_ACK);
476		goto dropnoprobe;
477	}
478drop:
479	TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
480dropnoprobe:
481	INP_WUNLOCK(inp);
482	m_freem(m);
483	return (0);
484}
485
486void
487tcp_twclose(struct tcptw *tw, int reuse)
488{
489	struct socket *so;
490	struct inpcb *inp;
491
492	/*
493	 * At this point, we are in one of two situations:
494	 *
495	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
496	 *     all state.
497	 *
498	 * (2) We have a socket -- if we own a reference, release it and
499	 *     notify the socket layer.
500	 */
501	inp = tw->tw_inpcb;
502	KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
503	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
504	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);	/* in_pcbfree() */
505	INP_WLOCK_ASSERT(inp);
506
507	tcp_tw_2msl_stop(tw, reuse);
508	inp->inp_ppcb = NULL;
509	in_pcbdrop(inp);
510
511	so = inp->inp_socket;
512	if (so != NULL) {
513		/*
514		 * If there's a socket, handle two cases: first, we own a
515		 * strong reference, which we will now release, or we don't
516		 * in which case another reference exists (XXXRW: think
517		 * about this more), and we don't need to take action.
518		 */
519		if (inp->inp_flags & INP_SOCKREF) {
520			inp->inp_flags &= ~INP_SOCKREF;
521			INP_WUNLOCK(inp);
522			ACCEPT_LOCK();
523			SOCK_LOCK(so);
524			KASSERT(so->so_state & SS_PROTOREF,
525			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
526			so->so_state &= ~SS_PROTOREF;
527			sofree(so);
528		} else {
529			/*
530			 * If we don't own the only reference, the socket and
531			 * inpcb need to be left around to be handled by
532			 * tcp_usr_detach() later.
533			 */
534			INP_WUNLOCK(inp);
535		}
536	} else {
537		/*
538		 * The socket has been already cleaned-up for us, only free the
539		 * inpcb.
540		 */
541		in_pcbfree(inp);
542	}
543	TCPSTAT_INC(tcps_closed);
544}
545
546static int
547tcp_twrespond(struct tcptw *tw, int flags)
548{
549	struct inpcb *inp = tw->tw_inpcb;
550#if defined(INET6) || defined(INET)
551	struct tcphdr *th = NULL;
552#endif
553	struct mbuf *m;
554#ifdef INET
555	struct ip *ip = NULL;
556#endif
557	u_int hdrlen, optlen;
558	int error = 0;			/* Keep compiler happy */
559	struct tcpopt to;
560#ifdef INET6
561	struct ip6_hdr *ip6 = NULL;
562	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
563#endif
564	hdrlen = 0;                     /* Keep compiler happy */
565
566	INP_WLOCK_ASSERT(inp);
567
568	m = m_gethdr(M_NOWAIT, MT_DATA);
569	if (m == NULL)
570		return (ENOBUFS);
571	m->m_data += max_linkhdr;
572
573#ifdef MAC
574	mac_inpcb_create_mbuf(inp, m);
575#endif
576
577#ifdef INET6
578	if (isipv6) {
579		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
580		ip6 = mtod(m, struct ip6_hdr *);
581		th = (struct tcphdr *)(ip6 + 1);
582		tcpip_fillheaders(inp, ip6, th);
583	}
584#endif
585#if defined(INET6) && defined(INET)
586	else
587#endif
588#ifdef INET
589	{
590		hdrlen = sizeof(struct tcpiphdr);
591		ip = mtod(m, struct ip *);
592		th = (struct tcphdr *)(ip + 1);
593		tcpip_fillheaders(inp, ip, th);
594	}
595#endif
596	to.to_flags = 0;
597
598	/*
599	 * Send a timestamp and echo-reply if both our side and our peer
600	 * have sent timestamps in our SYN's and this is not a RST.
601	 */
602	if (tw->t_recent && flags == TH_ACK) {
603		to.to_flags |= TOF_TS;
604		to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
605		to.to_tsecr = tw->t_recent;
606	}
607	optlen = tcp_addoptions(&to, (u_char *)(th + 1));
608
609	m->m_len = hdrlen + optlen;
610	m->m_pkthdr.len = m->m_len;
611
612	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
613
614	th->th_seq = htonl(tw->snd_nxt);
615	th->th_ack = htonl(tw->rcv_nxt);
616	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
617	th->th_flags = flags;
618	th->th_win = htons(tw->last_win);
619
620	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
621#ifdef INET6
622	if (isipv6) {
623		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
624		th->th_sum = in6_cksum_pseudo(ip6,
625		    sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
626		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
627		TCP_PROBE5(send, NULL, NULL, ip6, NULL, th);
628		error = ip6_output(m, inp->in6p_outputopts, NULL,
629		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
630	}
631#endif
632#if defined(INET6) && defined(INET)
633	else
634#endif
635#ifdef INET
636	{
637		m->m_pkthdr.csum_flags = CSUM_TCP;
638		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
639		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
640		ip->ip_len = htons(m->m_pkthdr.len);
641		if (V_path_mtu_discovery)
642			ip->ip_off |= htons(IP_DF);
643		TCP_PROBE5(send, NULL, NULL, ip, NULL, th);
644		error = ip_output(m, inp->inp_options, NULL,
645		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
646		    NULL, inp);
647	}
648#endif
649	if (flags & TH_ACK)
650		TCPSTAT_INC(tcps_sndacks);
651	else
652		TCPSTAT_INC(tcps_sndctrl);
653	TCPSTAT_INC(tcps_sndtotal);
654	return (error);
655}
656
657static void
658tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
659{
660
661	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
662	INP_WLOCK_ASSERT(tw->tw_inpcb);
663
664	TW_WLOCK(V_tw_lock);
665	if (rearm)
666		TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
667	tw->tw_time = ticks + 2 * tcp_msl;
668	TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
669	TW_WUNLOCK(V_tw_lock);
670}
671
672static void
673tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
674{
675	struct ucred *cred;
676	struct inpcb *inp;
677	int released;
678
679	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
680
681	TW_WLOCK(V_tw_lock);
682	inp = tw->tw_inpcb;
683	tw->tw_inpcb = NULL;
684
685	TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
686	cred = tw->tw_cred;
687	tw->tw_cred = NULL;
688	TW_WUNLOCK(V_tw_lock);
689
690	if (cred != NULL)
691		crfree(cred);
692
693	released = in_pcbrele_wlocked(inp);
694	KASSERT(!released, ("%s: inp should not be released here", __func__));
695
696	if (!reuse)
697		uma_zfree(V_tcptw_zone, tw);
698	TCPSTATES_DEC(TCPS_TIME_WAIT);
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				if (__predict_true(tw == NULL)) {
745					INP_INFO_RUNLOCK(&V_tcbinfo);
746					continue;
747				} else {
748					/* This should not happen as in TIMEWAIT
749					 * state the inp should not be destroyed
750					 * before its tcptw. If INVARIANTS is
751					 * defined panic.
752					 */
753#ifdef INVARIANTS
754					panic("%s: Panic before an infinite "
755					    "loop: INP_TIMEWAIT && (INP_FREED "
756					    "|| inp last reference) && tw != "
757					    "NULL", __func__);
758#else
759					log(LOG_ERR, "%s: Avoid an infinite "
760					    "loop: INP_TIMEWAIT && (INP_FREED "
761					    "|| inp last reference) && tw != "
762					    "NULL", __func__);
763#endif
764					INP_INFO_RUNLOCK(&V_tcbinfo);
765					break;
766				}
767			}
768
769			if (tw == NULL) {
770				/* tcp_twclose() has already been called */
771				INP_WUNLOCK(inp);
772				INP_INFO_RUNLOCK(&V_tcbinfo);
773				continue;
774			}
775
776			tcp_twclose(tw, reuse);
777			INP_INFO_RUNLOCK(&V_tcbinfo);
778			if (reuse)
779			    return tw;
780		} else {
781			/* INP_INFO lock is busy, continue later. */
782			INP_WLOCK(inp);
783			if (!in_pcbrele_wlocked(inp))
784				INP_WUNLOCK(inp);
785			break;
786		}
787	}
788
789	return NULL;
790}
791