2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
37 #include "opt_tcpdebug.h"
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>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
51 #include <sys/syslog.h>
53 #include <sys/protosw.h>
54 #include <sys/random.h>
58 #include <net/route.h>
60 #include <net/if_var.h>
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>
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>
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>
84 #include <netinet6/tcp6_var.h>
86 #include <netinet/tcpip.h>
88 #include <netinet/tcp_debug.h>
91 #include <netinet6/ip6protosw.h>
94 #include <machine/in_cksum.h>
96 #include <security/mac/mac_framework.h>
98 static VNET_DEFINE(uma_zone_t, tcptw_zone);
99 #define V_tcptw_zone VNET(tcptw_zone)
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.
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
113 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
114 #define V_twq_2msl VNET(twq_2msl)
116 /* Global timewait lock */
117 static VNET_DEFINE(struct rwlock, tw_lock);
118 #define V_tw_lock VNET(tw_lock)
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)
131 static void tcp_tw_2msl_reset(struct tcptw *, int);
132 static void tcp_tw_2msl_stop(struct tcptw *, int);
133 static int tcp_twrespond(struct tcptw *, int);
136 tcptw_auto_size(void)
141 * Max out at half the ephemeral port range so that TIME_WAIT
142 * sockets don't tie up too many ephemeral ports.
144 if (V_ipport_lastauto > V_ipport_firstauto)
145 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
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));
153 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
158 new = tcptw_auto_size();
161 error = sysctl_handle_int(oidp, &new, 0, req);
162 if (error == 0 && req->newptr)
165 uma_zone_set_max(V_tcptw_zone, maxtcptw);
170 SYSCTL_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");
174 VNET_DEFINE(int, nolocaltimewait) = 0;
175 #define V_nolocaltimewait VNET(nolocaltimewait)
176 SYSCTL_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");
181 tcp_tw_zone_change(void)
185 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
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);
196 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
198 uma_zone_set_max(V_tcptw_zone, maxtcptw);
199 TAILQ_INIT(&V_twq_2msl);
200 TW_LOCK_INIT(V_tw_lock, "tcptw");
209 INP_INFO_RLOCK(&V_tcbinfo);
210 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
212 INP_INFO_RUNLOCK(&V_tcbinfo);
214 TW_LOCK_DESTROY(V_tw_lock);
215 uma_zdestroy(V_tcptw_zone);
220 * Move a TCP connection into TIME_WAIT state.
222 * inp is locked, and is unlocked before returning.
225 tcp_twstart(struct tcpcb *tp)
228 struct inpcb *inp = tp->t_inpcb;
233 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
236 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
237 INP_WLOCK_ASSERT(inp);
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"));
243 if (V_nolocaltimewait) {
247 error = in6_localaddr(&inp->in6p_faddr);
249 #if defined(INET6) && defined(INET)
253 error = in_localip(inp->inp_faddr);
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.
268 tcp_state_change(tp, TCPS_TIME_WAIT);
270 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
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.
277 * XXX: Check if it possible to always have enough room
278 * in advance based on guarantees provided by uma_zalloc().
280 tw = tcp_tw_2msl_scan(1);
289 * The tcptw will hold a reference on its inpcb until tcp_twclose
293 in_pcbref(inp); /* Reference from tw */
296 * Recover last window size sent.
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)
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);
310 * Set t_recent if timestamps are used on the connection.
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;
321 tw->snd_nxt = tp->snd_nxt;
322 tw->rcv_nxt = tp->rcv_nxt;
325 tw->t_starttime = tp->t_starttime;
330 * be used for fin-wait-2 state also, then we may need
331 * a ts_recent from the last segment.
333 acknow = tp->t_flags & TF_ACKNOW;
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.
340 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
341 * and might not be needed here any longer.
344 soisdisconnected(so);
345 tw->tw_cred = crhold(so->so_cred);
347 tw->tw_so_options = so->so_options;
350 tcp_twrespond(tw, TH_ACK);
352 inp->inp_flags |= INP_TIMEWAIT;
353 TCPSTATES_INC(TCPS_TIME_WAIT);
354 tcp_tw_2msl_reset(tw, 0);
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.
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;
367 so->so_state &= ~SS_PROTOREF;
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.
378 tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
379 struct mbuf *m, int tlen)
385 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
386 INP_WLOCK_ASSERT(inp);
389 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
390 * still present. This is undesirable, but temporarily necessary
391 * until we work out how to handle inpcb's who's timewait state has
398 thflags = th->th_flags;
401 * NOTE: for FIN_WAIT_2 (to be added later),
402 * must validate sequence number before accepting RST
406 * If the segment contains RST:
407 * Drop the segment - see Stevens, vol. 2, p. 964 and
410 if (thflags & TH_RST)
414 /* PAWS not needed at the moment */
416 * RFC 1323 PAWS: If we have a timestamp reply on this segment
417 * and it's less than ts_recent, drop it.
419 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
420 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
421 if ((thflags & TH_ACK) == 0)
426 * ts_recent is never updated because we never accept new segments.
431 * If a new connection request is received
432 * while in TIME_WAIT, drop the old connection
433 * and start over if the sequence numbers
434 * are above the previous ones.
436 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
442 * Drop the segment if it does not contain an ACK.
444 if ((thflags & TH_ACK) == 0)
448 * Reset the 2MSL timer if this is a duplicate FIN.
450 if (thflags & TH_FIN) {
451 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
452 if (seq + 1 == tw->rcv_nxt)
453 tcp_tw_2msl_reset(tw, 1);
457 * Acknowledge the segment if it has data or is not a duplicate ACK.
459 if (thflags != TH_ACK || tlen != 0 ||
460 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) {
461 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
462 tcp_twrespond(tw, TH_ACK);
466 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
474 tcp_twclose(struct tcptw *tw, int reuse)
480 * At this point, we are in one of two situations:
482 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
485 * (2) We have a socket -- if we own a reference, release it and
486 * notify the socket layer.
489 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
490 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
491 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */
492 INP_WLOCK_ASSERT(inp);
494 tcp_tw_2msl_stop(tw, reuse);
495 inp->inp_ppcb = NULL;
498 so = inp->inp_socket;
501 * If there's a socket, handle two cases: first, we own a
502 * strong reference, which we will now release, or we don't
503 * in which case another reference exists (XXXRW: think
504 * about this more), and we don't need to take action.
506 if (inp->inp_flags & INP_SOCKREF) {
507 inp->inp_flags &= ~INP_SOCKREF;
511 KASSERT(so->so_state & SS_PROTOREF,
512 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
513 so->so_state &= ~SS_PROTOREF;
517 * If we don't own the only reference, the socket and
518 * inpcb need to be left around to be handled by
519 * tcp_usr_detach() later.
525 * The socket has been already cleaned-up for us, only free the
530 TCPSTAT_INC(tcps_closed);
534 tcp_twrespond(struct tcptw *tw, int flags)
536 struct inpcb *inp = tw->tw_inpcb;
537 #if defined(INET6) || defined(INET)
538 struct tcphdr *th = NULL;
542 struct ip *ip = NULL;
544 u_int hdrlen, optlen;
545 int error = 0; /* Keep compiler happy */
548 struct ip6_hdr *ip6 = NULL;
549 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
551 hdrlen = 0; /* Keep compiler happy */
553 INP_WLOCK_ASSERT(inp);
555 m = m_gethdr(M_NOWAIT, MT_DATA);
558 m->m_data += max_linkhdr;
561 mac_inpcb_create_mbuf(inp, m);
566 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
567 ip6 = mtod(m, struct ip6_hdr *);
568 th = (struct tcphdr *)(ip6 + 1);
569 tcpip_fillheaders(inp, ip6, th);
572 #if defined(INET6) && defined(INET)
577 hdrlen = sizeof(struct tcpiphdr);
578 ip = mtod(m, struct ip *);
579 th = (struct tcphdr *)(ip + 1);
580 tcpip_fillheaders(inp, ip, th);
586 * Send a timestamp and echo-reply if both our side and our peer
587 * have sent timestamps in our SYN's and this is not a RST.
589 if (tw->t_recent && flags == TH_ACK) {
590 to.to_flags |= TOF_TS;
591 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
592 to.to_tsecr = tw->t_recent;
594 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
596 m->m_len = hdrlen + optlen;
597 m->m_pkthdr.len = m->m_len;
599 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
601 th->th_seq = htonl(tw->snd_nxt);
602 th->th_ack = htonl(tw->rcv_nxt);
603 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
604 th->th_flags = flags;
605 th->th_win = htons(tw->last_win);
607 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
610 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
611 th->th_sum = in6_cksum_pseudo(ip6,
612 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
613 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
614 TCP_PROBE5(send, NULL, NULL, ip6, NULL, th);
615 error = ip6_output(m, inp->in6p_outputopts, NULL,
616 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
619 #if defined(INET6) && defined(INET)
624 m->m_pkthdr.csum_flags = CSUM_TCP;
625 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
626 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
627 ip->ip_len = htons(m->m_pkthdr.len);
628 if (V_path_mtu_discovery)
629 ip->ip_off |= htons(IP_DF);
630 TCP_PROBE5(send, NULL, NULL, ip, NULL, th);
631 error = ip_output(m, inp->inp_options, NULL,
632 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
637 TCPSTAT_INC(tcps_sndacks);
639 TCPSTAT_INC(tcps_sndctrl);
640 TCPSTAT_INC(tcps_sndtotal);
645 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
648 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
649 INP_WLOCK_ASSERT(tw->tw_inpcb);
653 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
654 tw->tw_time = ticks + 2 * tcp_msl;
655 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
656 TW_WUNLOCK(V_tw_lock);
660 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
666 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
672 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
675 TW_WUNLOCK(V_tw_lock);
680 released = in_pcbrele_wlocked(inp);
681 KASSERT(!released, ("%s: inp should not be released here", __func__));
684 uma_zfree(V_tcptw_zone, tw);
685 TCPSTATES_DEC(TCPS_TIME_WAIT);
689 tcp_tw_2msl_scan(int reuse)
697 * Exclusive pcbinfo lock is not required in reuse case even if
698 * two inpcb locks can be acquired simultaneously:
699 * - the inpcb transitioning to TIME_WAIT state in
701 * - the inpcb closed by tcp_twclose().
703 * It is because only inpcbs in FIN_WAIT2 or CLOSING states can
704 * transition in TIME_WAIT state. Then a pcbcb cannot be in
705 * TIME_WAIT list and transitioning to TIME_WAIT state at same
708 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
714 tw = TAILQ_FIRST(&V_twq_2msl);
715 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
716 TW_RUNLOCK(V_tw_lock);
719 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
724 TW_RUNLOCK(V_tw_lock);
726 if (INP_INFO_TRY_RLOCK(&V_tcbinfo)) {
730 if (in_pcbrele_wlocked(inp)) {
731 if (__predict_true(tw == NULL)) {
732 INP_INFO_RUNLOCK(&V_tcbinfo);
735 /* This should not happen as in TIMEWAIT
736 * state the inp should not be destroyed
737 * before its tcptw. If INVARIANTS is
741 panic("%s: Panic before an infinite "
742 "loop: INP_TIMEWAIT && (INP_FREED "
743 "|| inp last reference) && tw != "
746 log(LOG_ERR, "%s: Avoid an infinite "
747 "loop: INP_TIMEWAIT && (INP_FREED "
748 "|| inp last reference) && tw != "
751 INP_INFO_RUNLOCK(&V_tcbinfo);
757 /* tcp_twclose() has already been called */
759 INP_INFO_RUNLOCK(&V_tcbinfo);
763 tcp_twclose(tw, reuse);
764 INP_INFO_RUNLOCK(&V_tcbinfo);
768 /* INP_INFO lock is busy, continue later. */
770 if (!in_pcbrele_wlocked(inp))