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_pcb.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_icmp.h>
71 #include <netinet/ip_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/in6_pcb.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/scope6_var.h>
77 #include <netinet6/nd6.h>
79 #include <netinet/tcp.h>
80 #include <netinet/tcp_fsm.h>
81 #include <netinet/tcp_seq.h>
82 #include <netinet/tcp_timer.h>
83 #include <netinet/tcp_var.h>
85 #include <netinet6/tcp6_var.h>
87 #include <netinet/tcpip.h>
89 #include <netinet/tcp_debug.h>
92 #include <netinet6/ip6protosw.h>
95 #include <machine/in_cksum.h>
97 #include <security/mac/mac_framework.h>
99 static VNET_DEFINE(uma_zone_t, tcptw_zone);
100 #define V_tcptw_zone VNET(tcptw_zone)
104 * The timed wait queue contains references to each of the TCP sessions
105 * currently in the TIME_WAIT state. The queue pointers, including the
106 * queue pointers in each tcptw structure, are protected using the global
107 * timewait lock, which must be held over queue iteration and modification.
109 * Rules on tcptw usage:
110 * - a inpcb is always freed _after_ its tcptw
111 * - a tcptw relies on its inpcb reference counting for memory stability
112 * - a tcptw is dereferenceable only while its inpcb is locked
114 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
115 #define V_twq_2msl VNET(twq_2msl)
117 /* Global timewait lock */
118 static VNET_DEFINE(struct rwlock, tw_lock);
119 #define V_tw_lock VNET(tw_lock)
121 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0)
122 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw))
123 #define TW_RLOCK(tw) rw_rlock(&(tw))
124 #define TW_WLOCK(tw) rw_wlock(&(tw))
125 #define TW_RUNLOCK(tw) rw_runlock(&(tw))
126 #define TW_WUNLOCK(tw) rw_wunlock(&(tw))
127 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED)
128 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED)
129 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED)
130 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED)
132 static void tcp_tw_2msl_reset(struct tcptw *, int);
133 static void tcp_tw_2msl_stop(struct tcptw *, int);
134 static int tcp_twrespond(struct tcptw *, int);
137 tcptw_auto_size(void)
142 * Max out at half the ephemeral port range so that TIME_WAIT
143 * sockets don't tie up too many ephemeral ports.
145 if (V_ipport_lastauto > V_ipport_firstauto)
146 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
148 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
149 /* Protect against goofy port ranges smaller than 32. */
150 return (imin(imax(halfrange, 32), maxsockets / 5));
154 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
159 new = tcptw_auto_size();
162 error = sysctl_handle_int(oidp, &new, 0, req);
163 if (error == 0 && req->newptr)
166 uma_zone_set_max(V_tcptw_zone, maxtcptw);
171 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
172 &maxtcptw, 0, sysctl_maxtcptw, "IU",
173 "Maximum number of compressed TCP TIME_WAIT entries");
175 static VNET_DEFINE(int, nolocaltimewait) = 0;
176 #define V_nolocaltimewait VNET(nolocaltimewait)
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW,
178 &VNET_NAME(nolocaltimewait), 0,
179 "Do not create compressed TCP TIME_WAIT entries for local connections");
182 tcp_tw_zone_change(void)
186 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
193 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
194 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
195 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
197 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
199 uma_zone_set_max(V_tcptw_zone, maxtcptw);
200 TAILQ_INIT(&V_twq_2msl);
201 TW_LOCK_INIT(V_tw_lock, "tcptw");
209 struct epoch_tracker et;
211 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
212 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
214 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
216 TW_LOCK_DESTROY(V_tw_lock);
217 uma_zdestroy(V_tcptw_zone);
222 * Move a TCP connection into TIME_WAIT state.
224 * inp is locked, and is unlocked before returning.
227 tcp_twstart(struct tcpcb *tp)
229 struct tcptw twlocal, *tw;
230 struct inpcb *inp = tp->t_inpcb;
234 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
237 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
238 INP_WLOCK_ASSERT(inp);
240 /* A dropped inp should never transition to TIME_WAIT state. */
241 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
242 "(inp->inp_flags & INP_DROPPED) != 0"));
244 if (V_nolocaltimewait) {
247 local = in6_localaddr(&inp->in6p_faddr);
251 local = in_localip(inp->inp_faddr);
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.
262 tcp_state_change(tp, TCPS_TIME_WAIT);
267 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
270 * Reached limit on total number of TIMEWAIT connections
271 * allowed. Remove a connection from TIMEWAIT queue in LRU
272 * fashion to make room for this connection.
274 * XXX: Check if it possible to always have enough room
275 * in advance based on guarantees provided by uma_zalloc().
277 tw = tcp_tw_2msl_scan(1);
286 * For !local case the tcptw will hold a reference on its inpcb
287 * until tcp_twclose is called.
292 * Recover last window size sent.
294 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
295 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
300 * Set t_recent if timestamps are used on the connection.
302 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
303 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
304 tw->t_recent = tp->ts_recent;
305 tw->ts_offset = tp->ts_offset;
311 tw->snd_nxt = tp->snd_nxt;
312 tw->rcv_nxt = tp->rcv_nxt;
315 tw->t_starttime = tp->t_starttime;
320 * be used for fin-wait-2 state also, then we may need
321 * a ts_recent from the last segment.
323 acknow = tp->t_flags & TF_ACKNOW;
326 * First, discard tcpcb state, which includes stopping its timers and
327 * freeing it. tcp_discardcb() used to also release the inpcb, but
328 * that work is now done in the caller.
330 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
331 * and might not be needed here any longer.
334 so = inp->inp_socket;
335 soisdisconnected(so);
336 tw->tw_so_options = so->so_options;
337 inp->inp_flags |= INP_TIMEWAIT;
339 tcp_twrespond(tw, TH_ACK);
343 in_pcbref(inp); /* Reference from tw */
344 tw->tw_cred = crhold(so->so_cred);
346 TCPSTATES_INC(TCPS_TIME_WAIT);
347 tcp_tw_2msl_reset(tw, 0);
351 * If the inpcb owns the sole reference to the socket, then we can
352 * detach and free the socket as it is not needed in time wait.
354 if (inp->inp_flags & INP_SOCKREF) {
355 KASSERT(so->so_state & SS_PROTOREF,
356 ("tcp_twstart: !SS_PROTOREF"));
357 inp->inp_flags &= ~INP_SOCKREF;
360 so->so_state &= ~SS_PROTOREF;
367 * Returns 1 if the TIME_WAIT state was killed and we should start over,
368 * looking for a pcb in the listen state. Returns 0 otherwise.
371 tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
372 struct mbuf *m, int tlen)
378 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
379 INP_WLOCK_ASSERT(inp);
382 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
383 * still present. This is undesirable, but temporarily necessary
384 * until we work out how to handle inpcb's who's timewait state has
391 thflags = th->th_flags;
394 * NOTE: for FIN_WAIT_2 (to be added later),
395 * must validate sequence number before accepting RST
399 * If the segment contains RST:
400 * Drop the segment - see Stevens, vol. 2, p. 964 and
403 if (thflags & TH_RST)
407 /* PAWS not needed at the moment */
409 * RFC 1323 PAWS: If we have a timestamp reply on this segment
410 * and it's less than ts_recent, drop it.
412 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
413 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
414 if ((thflags & TH_ACK) == 0)
419 * ts_recent is never updated because we never accept new segments.
424 * If a new connection request is received
425 * while in TIME_WAIT, drop the old connection
426 * and start over if the sequence numbers
427 * are above the previous ones.
429 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
435 * Drop the segment if it does not contain an ACK.
437 if ((thflags & TH_ACK) == 0)
441 * Reset the 2MSL timer if this is a duplicate FIN.
443 if (thflags & TH_FIN) {
444 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
445 if (seq + 1 == tw->rcv_nxt)
446 tcp_tw_2msl_reset(tw, 1);
450 * Acknowledge the segment if it has data or is not a duplicate ACK.
452 if (thflags != TH_ACK || tlen != 0 ||
453 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
454 tcp_twrespond(tw, TH_ACK);
462 tcp_twclose(struct tcptw *tw, int reuse)
468 * At this point, we are in one of two situations:
470 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
473 * (2) We have a socket -- if we own a reference, release it and
474 * notify the socket layer.
477 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
478 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
479 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */
480 INP_WLOCK_ASSERT(inp);
482 tcp_tw_2msl_stop(tw, reuse);
483 inp->inp_ppcb = NULL;
486 so = inp->inp_socket;
489 * If there's a socket, handle two cases: first, we own a
490 * strong reference, which we will now release, or we don't
491 * in which case another reference exists (XXXRW: think
492 * about this more), and we don't need to take action.
494 if (inp->inp_flags & INP_SOCKREF) {
495 inp->inp_flags &= ~INP_SOCKREF;
498 KASSERT(so->so_state & SS_PROTOREF,
499 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
500 so->so_state &= ~SS_PROTOREF;
504 * If we don't own the only reference, the socket and
505 * inpcb need to be left around to be handled by
506 * tcp_usr_detach() later.
512 * The socket has been already cleaned-up for us, only free the
517 TCPSTAT_INC(tcps_closed);
521 tcp_twrespond(struct tcptw *tw, int flags)
523 struct inpcb *inp = tw->tw_inpcb;
524 #if defined(INET6) || defined(INET)
525 struct tcphdr *th = NULL;
529 struct ip *ip = NULL;
531 u_int hdrlen, optlen;
532 int error = 0; /* Keep compiler happy */
535 struct ip6_hdr *ip6 = NULL;
536 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
538 hdrlen = 0; /* Keep compiler happy */
540 INP_WLOCK_ASSERT(inp);
542 m = m_gethdr(M_NOWAIT, MT_DATA);
545 m->m_data += max_linkhdr;
548 mac_inpcb_create_mbuf(inp, m);
553 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
554 ip6 = mtod(m, struct ip6_hdr *);
555 th = (struct tcphdr *)(ip6 + 1);
556 tcpip_fillheaders(inp, ip6, th);
559 #if defined(INET6) && defined(INET)
564 hdrlen = sizeof(struct tcpiphdr);
565 ip = mtod(m, struct ip *);
566 th = (struct tcphdr *)(ip + 1);
567 tcpip_fillheaders(inp, ip, th);
573 * Send a timestamp and echo-reply if both our side and our peer
574 * have sent timestamps in our SYN's and this is not a RST.
576 if (tw->t_recent && flags == TH_ACK) {
577 to.to_flags |= TOF_TS;
578 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
579 to.to_tsecr = tw->t_recent;
581 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
583 m->m_len = hdrlen + optlen;
584 m->m_pkthdr.len = m->m_len;
586 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
588 th->th_seq = htonl(tw->snd_nxt);
589 th->th_ack = htonl(tw->rcv_nxt);
590 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
591 th->th_flags = flags;
592 th->th_win = htons(tw->last_win);
594 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
597 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
598 th->th_sum = in6_cksum_pseudo(ip6,
599 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
600 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
601 error = ip6_output(m, inp->in6p_outputopts, NULL,
602 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
605 #if defined(INET6) && defined(INET)
610 m->m_pkthdr.csum_flags = CSUM_TCP;
611 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
612 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
613 ip->ip_len = htons(m->m_pkthdr.len);
614 if (V_path_mtu_discovery)
615 ip->ip_off |= htons(IP_DF);
616 error = ip_output(m, inp->inp_options, NULL,
617 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
622 TCPSTAT_INC(tcps_sndacks);
624 TCPSTAT_INC(tcps_sndctrl);
625 TCPSTAT_INC(tcps_sndtotal);
630 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
633 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
634 INP_WLOCK_ASSERT(tw->tw_inpcb);
638 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
639 tw->tw_time = ticks + 2 * tcp_msl;
640 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
641 TW_WUNLOCK(V_tw_lock);
645 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
649 int released __unused;
651 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
657 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
660 TW_WUNLOCK(V_tw_lock);
665 released = in_pcbrele_wlocked(inp);
666 KASSERT(!released, ("%s: inp should not be released here", __func__));
669 uma_zfree(V_tcptw_zone, tw);
670 TCPSTATES_DEC(TCPS_TIME_WAIT);
674 tcp_tw_2msl_scan(int reuse)
678 struct epoch_tracker et;
683 * Exclusive pcbinfo lock is not required in reuse case even if
684 * two inpcb locks can be acquired simultaneously:
685 * - the inpcb transitioning to TIME_WAIT state in
687 * - the inpcb closed by tcp_twclose().
689 * It is because only inpcbs in FIN_WAIT2 or CLOSING states can
690 * transition in TIME_WAIT state. Then a pcbcb cannot be in
691 * TIME_WAIT list and transitioning to TIME_WAIT state at same
694 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
700 tw = TAILQ_FIRST(&V_twq_2msl);
701 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
702 TW_RUNLOCK(V_tw_lock);
705 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
710 TW_RUNLOCK(V_tw_lock);
712 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
715 if (in_pcbrele_wlocked(inp)) {
716 if (__predict_true(tw == NULL)) {
717 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
720 /* This should not happen as in TIMEWAIT
721 * state the inp should not be destroyed
722 * before its tcptw. If INVARIANTS is
726 panic("%s: Panic before an infinite "
727 "loop: INP_TIMEWAIT && (INP_FREED "
728 "|| inp last reference) && tw != "
731 log(LOG_ERR, "%s: Avoid an infinite "
732 "loop: INP_TIMEWAIT && (INP_FREED "
733 "|| inp last reference) && tw != "
736 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
742 /* tcp_twclose() has already been called */
744 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
748 tcp_twclose(tw, reuse);
749 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);