2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
53 #include <sys/syslog.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
60 #include <net/route.h>
62 #include <net/if_var.h>
65 #include <netinet/in.h>
66 #include <netinet/in_kdtrace.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip_icmp.h>
72 #include <netinet/ip_var.h>
74 #include <netinet/ip6.h>
75 #include <netinet6/in6_pcb.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
80 #include <netinet/tcp.h>
81 #include <netinet/tcp_fsm.h>
82 #include <netinet/tcp_seq.h>
83 #include <netinet/tcp_timer.h>
84 #include <netinet/tcp_var.h>
86 #include <netinet6/tcp6_var.h>
88 #include <netinet/tcpip.h>
90 #include <netinet/tcp_debug.h>
93 #include <netinet6/ip6protosw.h>
96 #include <netinet/udp.h>
97 #include <netinet/udp_var.h>
98 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
102 VNET_DEFINE_STATIC(uma_zone_t, tcptw_zone);
103 #define V_tcptw_zone VNET(tcptw_zone)
107 * The timed wait queue contains references to each of the TCP sessions
108 * currently in the TIME_WAIT state. The queue pointers, including the
109 * queue pointers in each tcptw structure, are protected using the global
110 * timewait lock, which must be held over queue iteration and modification.
112 * Rules on tcptw usage:
113 * - a inpcb is always freed _after_ its tcptw
114 * - a tcptw relies on its inpcb reference counting for memory stability
115 * - a tcptw is dereferenceable only while its inpcb is locked
117 VNET_DEFINE_STATIC(TAILQ_HEAD(, tcptw), twq_2msl);
118 #define V_twq_2msl VNET(twq_2msl)
120 /* Global timewait lock */
121 VNET_DEFINE_STATIC(struct rwlock, tw_lock);
122 #define V_tw_lock VNET(tw_lock)
124 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0)
125 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw))
126 #define TW_RLOCK(tw) rw_rlock(&(tw))
127 #define TW_WLOCK(tw) rw_wlock(&(tw))
128 #define TW_RUNLOCK(tw) rw_runlock(&(tw))
129 #define TW_WUNLOCK(tw) rw_wunlock(&(tw))
130 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED)
131 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED)
132 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED)
133 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED)
135 static void tcp_tw_2msl_reset(struct tcptw *, int);
136 static void tcp_tw_2msl_stop(struct tcptw *, int);
137 static int tcp_twrespond(struct tcptw *, int);
140 tcptw_auto_size(void)
145 * Max out at half the ephemeral port range so that TIME_WAIT
146 * sockets don't tie up too many ephemeral ports.
148 if (V_ipport_lastauto > V_ipport_firstauto)
149 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
151 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
152 /* Protect against goofy port ranges smaller than 32. */
153 return (imin(imax(halfrange, 32), maxsockets / 5));
157 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
162 new = tcptw_auto_size();
165 error = sysctl_handle_int(oidp, &new, 0, req);
166 if (error == 0 && req->newptr)
169 uma_zone_set_max(V_tcptw_zone, maxtcptw);
174 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw,
175 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
176 &maxtcptw, 0, sysctl_maxtcptw, "IU",
177 "Maximum number of compressed TCP TIME_WAIT entries");
179 VNET_DEFINE_STATIC(int, nolocaltimewait) = 0;
180 #define V_nolocaltimewait VNET(nolocaltimewait)
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW,
182 &VNET_NAME(nolocaltimewait), 0,
183 "Do not create compressed TCP TIME_WAIT entries for local connections");
186 tcp_tw_zone_change(void)
190 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
197 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
199 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
201 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
203 uma_zone_set_max(V_tcptw_zone, maxtcptw);
204 TAILQ_INIT(&V_twq_2msl);
205 TW_LOCK_INIT(V_tw_lock, "tcptw");
213 struct epoch_tracker et;
216 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
220 TW_LOCK_DESTROY(V_tw_lock);
221 uma_zdestroy(V_tcptw_zone);
226 * Move a TCP connection into TIME_WAIT state.
228 * inp is locked, and is unlocked before returning.
231 tcp_twstart(struct tcpcb *tp)
233 struct tcptw twlocal, *tw;
234 struct inpcb *inp = tp->t_inpcb;
239 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
243 INP_WLOCK_ASSERT(inp);
245 /* A dropped inp should never transition to TIME_WAIT state. */
246 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
247 "(inp->inp_flags & INP_DROPPED) != 0"));
249 if (V_nolocaltimewait) {
252 local = in6_localaddr(&inp->in6p_faddr);
256 local = in_localip(inp->inp_faddr);
264 * For use only by DTrace. We do not reference the state
265 * after this point so modifying it in place is not a problem.
267 tcp_state_change(tp, TCPS_TIME_WAIT);
272 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
275 * Reached limit on total number of TIMEWAIT connections
276 * allowed. Remove a connection from TIMEWAIT queue in LRU
277 * fashion to make room for this connection.
279 * XXX: Check if it possible to always have enough room
280 * in advance based on guarantees provided by uma_zalloc().
282 tw = tcp_tw_2msl_scan(1);
291 * For !local case the tcptw will hold a reference on its inpcb
292 * until tcp_twclose is called.
297 * Recover last window size sent.
299 so = inp->inp_socket;
300 recwin = lmin(lmax(sbspace(&so->so_rcv), 0),
301 (long)TCP_MAXWIN << tp->rcv_scale);
302 if (recwin < (so->so_rcv.sb_hiwat / 4) &&
303 recwin < tp->t_maxseg)
305 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
306 recwin < (tp->rcv_adv - tp->rcv_nxt))
307 recwin = (tp->rcv_adv - tp->rcv_nxt);
308 tw->last_win = (u_short)(recwin >> tp->rcv_scale);
311 * Set t_recent if timestamps are used on the connection.
313 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
314 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
315 tw->t_recent = tp->ts_recent;
316 tw->ts_offset = tp->ts_offset;
322 tw->snd_nxt = tp->snd_nxt;
323 tw->t_port = tp->t_port;
324 tw->rcv_nxt = tp->rcv_nxt;
327 tw->t_starttime = tp->t_starttime;
332 * be used for fin-wait-2 state also, then we may need
333 * a ts_recent from the last segment.
335 acknow = tp->t_flags & TF_ACKNOW;
338 * First, discard tcpcb state, which includes stopping its timers and
339 * freeing it. tcp_discardcb() used to also release the inpcb, but
340 * that work is now done in the caller.
342 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
343 * and might not be needed here any longer.
346 soisdisconnected(so);
347 tw->tw_so_options = so->so_options;
348 inp->inp_flags |= INP_TIMEWAIT;
350 tcp_twrespond(tw, TH_ACK);
354 in_pcbref(inp); /* Reference from tw */
355 tw->tw_cred = crhold(so->so_cred);
357 TCPSTATES_INC(TCPS_TIME_WAIT);
358 tcp_tw_2msl_reset(tw, 0);
362 * If the inpcb owns the sole reference to the socket, then we can
363 * detach and free the socket as it is not needed in time wait.
365 if (inp->inp_flags & INP_SOCKREF) {
366 KASSERT(so->so_state & SS_PROTOREF,
367 ("tcp_twstart: !SS_PROTOREF"));
368 inp->inp_flags &= ~INP_SOCKREF;
371 so->so_state &= ~SS_PROTOREF;
378 * Returns 1 if the TIME_WAIT state was killed and we should start over,
379 * looking for a pcb in the listen state. Returns 0 otherwise.
380 * It be called with to == NULL only for pure SYN-segments.
383 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
384 struct mbuf *m, int tlen)
391 INP_WLOCK_ASSERT(inp);
394 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
395 * still present. This is undesirable, but temporarily necessary
396 * until we work out how to handle inpcb's who's timewait state has
403 thflags = th->th_flags;
404 KASSERT(to != NULL || (thflags & (TH_SYN | TH_ACK)) == TH_SYN,
405 ("tcp_twcheck: called without options on a non-SYN segment"));
408 * NOTE: for FIN_WAIT_2 (to be added later),
409 * must validate sequence number before accepting RST
413 * If the segment contains RST:
414 * Drop the segment - see Stevens, vol. 2, p. 964 and
417 if (thflags & TH_RST)
421 /* PAWS not needed at the moment */
423 * RFC 1323 PAWS: If we have a timestamp reply on this segment
424 * and it's less than ts_recent, drop it.
426 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
427 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
428 if ((thflags & TH_ACK) == 0)
433 * ts_recent is never updated because we never accept new segments.
438 * If a new connection request is received
439 * while in TIME_WAIT, drop the old connection
440 * and start over if the sequence numbers
441 * are above the previous ones.
442 * Allow UDP port number changes in this case.
444 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
450 * Send RST if UDP port numbers don't match
452 if (tw->t_port != m->m_pkthdr.tcp_tun_port) {
453 if (th->th_flags & TH_ACK) {
454 tcp_respond(NULL, mtod(m, void *), th, m,
455 (tcp_seq)0, th->th_ack, TH_RST);
457 if (th->th_flags & TH_SYN)
459 if (th->th_flags & TH_FIN)
461 tcp_respond(NULL, mtod(m, void *), th, m,
462 th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK);
469 * Drop the segment if it does not contain an ACK.
471 if ((thflags & TH_ACK) == 0)
475 * If timestamps were negotiated during SYN/ACK and a
476 * segment without a timestamp is received, silently drop
477 * the segment, unless the missing timestamps are tolerated.
478 * See section 3.2 of RFC 7323.
480 if (((to->to_flags & TOF_TS) == 0) && (tw->t_recent != 0) &&
481 (V_tcp_tolerate_missing_ts == 0)) {
486 * Reset the 2MSL timer if this is a duplicate FIN.
488 if (thflags & TH_FIN) {
489 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
490 if (seq + 1 == tw->rcv_nxt)
491 tcp_tw_2msl_reset(tw, 1);
495 * Acknowledge the segment if it has data or is not a duplicate ACK.
497 if (thflags != TH_ACK || tlen != 0 ||
498 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) {
499 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
500 tcp_twrespond(tw, TH_ACK);
504 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
512 tcp_twclose(struct tcptw *tw, int reuse)
518 * At this point, we are in one of two situations:
520 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
523 * (2) We have a socket -- if we own a reference, release it and
524 * notify the socket layer.
527 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
528 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
530 INP_WLOCK_ASSERT(inp);
532 tcp_tw_2msl_stop(tw, reuse);
533 inp->inp_ppcb = NULL;
536 so = inp->inp_socket;
539 * If there's a socket, handle two cases: first, we own a
540 * strong reference, which we will now release, or we don't
541 * in which case another reference exists (XXXRW: think
542 * about this more), and we don't need to take action.
544 if (inp->inp_flags & INP_SOCKREF) {
545 inp->inp_flags &= ~INP_SOCKREF;
548 KASSERT(so->so_state & SS_PROTOREF,
549 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
550 so->so_state &= ~SS_PROTOREF;
554 * If we don't own the only reference, the socket and
555 * inpcb need to be left around to be handled by
556 * tcp_usr_detach() later.
562 * The socket has been already cleaned-up for us, only free the
567 TCPSTAT_INC(tcps_closed);
571 tcp_twrespond(struct tcptw *tw, int flags)
573 struct inpcb *inp = tw->tw_inpcb;
574 #if defined(INET6) || defined(INET)
575 struct tcphdr *th = NULL;
579 struct ip *ip = NULL;
581 u_int hdrlen, optlen, ulen;
582 int error = 0; /* Keep compiler happy */
585 struct ip6_hdr *ip6 = NULL;
586 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
588 struct udphdr *udp = NULL;
589 hdrlen = 0; /* Keep compiler happy */
591 INP_WLOCK_ASSERT(inp);
593 m = m_gethdr(M_NOWAIT, MT_DATA);
596 m->m_data += max_linkhdr;
599 mac_inpcb_create_mbuf(inp, m);
604 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
605 ip6 = mtod(m, struct ip6_hdr *);
607 udp = (struct udphdr *)(ip6 + 1);
608 hdrlen += sizeof(struct udphdr);
609 udp->uh_sport = htons(V_tcp_udp_tunneling_port);
610 udp->uh_dport = tw->t_port;
611 ulen = (hdrlen - sizeof(struct ip6_hdr));
612 th = (struct tcphdr *)(udp + 1);
614 th = (struct tcphdr *)(ip6 + 1);
615 tcpip_fillheaders(inp, tw->t_port, ip6, th);
618 #if defined(INET6) && defined(INET)
623 hdrlen = sizeof(struct tcpiphdr);
624 ip = mtod(m, struct ip *);
626 udp = (struct udphdr *)(ip + 1);
627 hdrlen += sizeof(struct udphdr);
628 udp->uh_sport = htons(V_tcp_udp_tunneling_port);
629 udp->uh_dport = tw->t_port;
630 ulen = (hdrlen - sizeof(struct ip));
631 th = (struct tcphdr *)(udp + 1);
633 th = (struct tcphdr *)(ip + 1);
634 tcpip_fillheaders(inp, tw->t_port, ip, th);
640 * Send a timestamp and echo-reply if both our side and our peer
641 * have sent timestamps in our SYN's and this is not a RST.
643 if (tw->t_recent && flags == TH_ACK) {
644 to.to_flags |= TOF_TS;
645 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
646 to.to_tsecr = tw->t_recent;
648 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
652 udp->uh_ulen = htons(ulen);
654 m->m_len = hdrlen + optlen;
655 m->m_pkthdr.len = m->m_len;
657 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
659 th->th_seq = htonl(tw->snd_nxt);
660 th->th_ack = htonl(tw->rcv_nxt);
661 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
662 th->th_flags = flags;
663 th->th_win = htons(tw->last_win);
668 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
669 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
670 udp->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
671 th->th_sum = htons(0);
673 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
674 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
675 th->th_sum = in6_cksum_pseudo(ip6,
676 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
678 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
679 TCP_PROBE5(send, NULL, NULL, ip6, NULL, th);
680 error = ip6_output(m, inp->in6p_outputopts, NULL,
681 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
684 #if defined(INET6) && defined(INET)
690 m->m_pkthdr.csum_flags = CSUM_UDP;
691 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
692 udp->uh_sum = in_pseudo(ip->ip_src.s_addr,
693 ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP));
694 th->th_sum = htons(0);
696 m->m_pkthdr.csum_flags = CSUM_TCP;
697 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
698 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
699 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
701 ip->ip_len = htons(m->m_pkthdr.len);
702 if (V_path_mtu_discovery)
703 ip->ip_off |= htons(IP_DF);
704 TCP_PROBE5(send, NULL, NULL, ip, NULL, th);
705 error = ip_output(m, inp->inp_options, NULL,
706 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
711 TCPSTAT_INC(tcps_sndacks);
713 TCPSTAT_INC(tcps_sndctrl);
714 TCPSTAT_INC(tcps_sndtotal);
719 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
723 INP_WLOCK_ASSERT(tw->tw_inpcb);
727 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
728 tw->tw_time = ticks + 2 * tcp_msl;
729 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
730 TW_WUNLOCK(V_tw_lock);
734 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
738 int released __unused;
746 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
749 TW_WUNLOCK(V_tw_lock);
754 released = in_pcbrele_wlocked(inp);
755 KASSERT(!released, ("%s: inp should not be released here", __func__));
758 uma_zfree(V_tcptw_zone, tw);
759 TCPSTATES_DEC(TCPS_TIME_WAIT);
763 tcp_tw_2msl_scan(int reuse)
772 tw = TAILQ_FIRST(&V_twq_2msl);
773 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
774 TW_RUNLOCK(V_tw_lock);
777 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
782 TW_RUNLOCK(V_tw_lock);
786 if (in_pcbrele_wlocked(inp)) {
787 if (__predict_true(tw == NULL)) {
790 /* This should not happen as in TIMEWAIT
791 * state the inp should not be destroyed
792 * before its tcptw. If INVARIANTS is
796 panic("%s: Panic before an infinite "
797 "loop: INP_TIMEWAIT && (INP_FREED "
798 "|| inp last reference) && tw != "
801 log(LOG_ERR, "%s: Avoid an infinite "
802 "loop: INP_TIMEWAIT && (INP_FREED "
803 "|| inp last reference) && tw != "
811 /* tcp_twclose() has already been called */
816 tcp_twclose(tw, reuse);