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>
50 #include <sys/protosw.h>
51 #include <sys/random.h>
55 #include <net/route.h>
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>
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>
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>
79 #include <netinet6/tcp6_var.h>
81 #include <netinet/tcpip.h>
83 #include <netinet/tcp_debug.h>
86 #include <netinet6/ip6protosw.h>
89 #include <machine/in_cksum.h>
91 #include <security/mac/mac_framework.h>
93 static VNET_DEFINE(uma_zone_t, tcptw_zone);
94 #define V_tcptw_zone VNET(tcptw_zone)
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.
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
108 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
109 #define V_twq_2msl VNET(twq_2msl)
111 /* Global timewait lock */
112 static VNET_DEFINE(struct rwlock, tw_lock);
113 #define V_tw_lock VNET(tw_lock)
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)
126 static void tcp_tw_2msl_reset(struct tcptw *, int);
127 static void tcp_tw_2msl_stop(struct tcptw *, int);
128 static int tcp_twrespond(struct tcptw *, int);
131 tcptw_auto_size(void)
136 * Max out at half the ephemeral port range so that TIME_WAIT
137 * sockets don't tie up too many ephemeral ports.
139 if (V_ipport_lastauto > V_ipport_firstauto)
140 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
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));
148 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
153 new = tcptw_auto_size();
156 error = sysctl_handle_int(oidp, &new, 0, req);
157 if (error == 0 && req->newptr)
160 uma_zone_set_max(V_tcptw_zone, maxtcptw);
165 SYSCTL_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");
169 VNET_DEFINE(int, nolocaltimewait) = 0;
170 #define V_nolocaltimewait VNET(nolocaltimewait)
171 SYSCTL_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");
176 tcp_tw_zone_change(void)
180 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
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);
191 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
193 uma_zone_set_max(V_tcptw_zone, maxtcptw);
194 TAILQ_INIT(&V_twq_2msl);
195 TW_LOCK_INIT(V_tw_lock, "tcptw");
204 INP_INFO_WLOCK(&V_tcbinfo);
205 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
207 INP_INFO_WUNLOCK(&V_tcbinfo);
209 TW_LOCK_DESTROY(V_tw_lock);
210 uma_zdestroy(V_tcptw_zone);
215 * Move a TCP connection into TIME_WAIT state.
217 * inp is locked, and is unlocked before returning.
220 tcp_twstart(struct tcpcb *tp)
223 struct inpcb *inp = tp->t_inpcb;
227 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
230 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
231 INP_WLOCK_ASSERT(inp);
233 if (V_nolocaltimewait) {
237 error = in6_localaddr(&inp->in6p_faddr);
239 #if defined(INET6) && defined(INET)
243 error = in_localip(inp->inp_faddr);
255 * For use only by DTrace. We do not reference the state
256 * after this point so modifying it in place is not a problem.
258 tcp_state_change(tp, TCPS_TIME_WAIT);
260 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
263 * Reached limit on total number of TIMEWAIT connections
264 * allowed. Remove a connection from TIMEWAIT queue in LRU
265 * fashion to make room for this connection.
267 * pcbinfo lock is needed here to prevent deadlock as
268 * two inpcb locks can be acquired simultaneously.
270 tw = tcp_tw_2msl_scan(1);
279 * The tcptw will hold a reference on its inpcb until tcp_twclose
283 in_pcbref(inp); /* Reference from tw */
286 * Recover last window size sent.
288 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
289 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
294 * Set t_recent if timestamps are used on the connection.
296 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
297 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
298 tw->t_recent = tp->ts_recent;
299 tw->ts_offset = tp->ts_offset;
305 tw->snd_nxt = tp->snd_nxt;
306 tw->rcv_nxt = tp->rcv_nxt;
309 tw->t_starttime = tp->t_starttime;
314 * be used for fin-wait-2 state also, then we may need
315 * a ts_recent from the last segment.
317 acknow = tp->t_flags & TF_ACKNOW;
320 * First, discard tcpcb state, which includes stopping its timers and
321 * freeing it. tcp_discardcb() used to also release the inpcb, but
322 * that work is now done in the caller.
324 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
325 * and might not be needed here any longer.
328 so = inp->inp_socket;
329 soisdisconnected(so);
330 tw->tw_cred = crhold(so->so_cred);
332 tw->tw_so_options = so->so_options;
335 tcp_twrespond(tw, TH_ACK);
337 inp->inp_flags |= INP_TIMEWAIT;
338 tcp_tw_2msl_reset(tw, 0);
341 * If the inpcb owns the sole reference to the socket, then we can
342 * detach and free the socket as it is not needed in time wait.
344 if (inp->inp_flags & INP_SOCKREF) {
345 KASSERT(so->so_state & SS_PROTOREF,
346 ("tcp_twstart: !SS_PROTOREF"));
347 inp->inp_flags &= ~INP_SOCKREF;
351 so->so_state &= ~SS_PROTOREF;
359 * The appromixate rate of ISN increase of Microsoft TCP stacks;
360 * the actual rate is slightly higher due to the addition of
361 * random positive increments.
363 * Most other new OSes use semi-randomized ISN values, so we
364 * do not need to worry about them.
366 #define MS_ISN_BYTES_PER_SECOND 250000
369 * Determine if the ISN we will generate has advanced beyond the last
370 * sequence number used by the previous connection. If so, indicate
371 * that it is safe to recycle this tw socket by returning 1.
374 tcp_twrecycleable(struct tcptw *tw)
376 tcp_seq new_iss = tw->iss;
377 tcp_seq new_irs = tw->irs;
379 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
380 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
381 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
383 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
391 * Returns 1 if the TIME_WAIT state was killed and we should start over,
392 * looking for a pcb in the listen state. Returns 0 otherwise.
395 tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
396 struct mbuf *m, int tlen)
402 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
403 INP_WLOCK_ASSERT(inp);
406 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
407 * still present. This is undesirable, but temporarily necessary
408 * until we work out how to handle inpcb's who's timewait state has
415 thflags = th->th_flags;
418 * NOTE: for FIN_WAIT_2 (to be added later),
419 * must validate sequence number before accepting RST
423 * If the segment contains RST:
424 * Drop the segment - see Stevens, vol. 2, p. 964 and
427 if (thflags & TH_RST)
431 /* PAWS not needed at the moment */
433 * RFC 1323 PAWS: If we have a timestamp reply on this segment
434 * and it's less than ts_recent, drop it.
436 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
437 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
438 if ((thflags & TH_ACK) == 0)
443 * ts_recent is never updated because we never accept new segments.
448 * If a new connection request is received
449 * while in TIME_WAIT, drop the old connection
450 * and start over if the sequence numbers
451 * are above the previous ones.
453 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
459 * Drop the segment if it does not contain an ACK.
461 if ((thflags & TH_ACK) == 0)
465 * Reset the 2MSL timer if this is a duplicate FIN.
467 if (thflags & TH_FIN) {
468 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
469 if (seq + 1 == tw->rcv_nxt)
470 tcp_tw_2msl_reset(tw, 1);
474 * Acknowledge the segment if it has data or is not a duplicate ACK.
476 if (thflags != TH_ACK || tlen != 0 ||
477 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
478 tcp_twrespond(tw, TH_ACK);
486 tcp_twclose(struct tcptw *tw, int reuse)
492 * At this point, we are in one of two situations:
494 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
497 * (2) We have a socket -- if we own a reference, release it and
498 * notify the socket layer.
501 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
502 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
503 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */
504 INP_WLOCK_ASSERT(inp);
506 tcp_tw_2msl_stop(tw, reuse);
507 inp->inp_ppcb = NULL;
510 so = inp->inp_socket;
513 * If there's a socket, handle two cases: first, we own a
514 * strong reference, which we will now release, or we don't
515 * in which case another reference exists (XXXRW: think
516 * about this more), and we don't need to take action.
518 if (inp->inp_flags & INP_SOCKREF) {
519 inp->inp_flags &= ~INP_SOCKREF;
523 KASSERT(so->so_state & SS_PROTOREF,
524 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
525 so->so_state &= ~SS_PROTOREF;
529 * If we don't own the only reference, the socket and
530 * inpcb need to be left around to be handled by
531 * tcp_usr_detach() later.
537 * The socket has been already cleaned-up for us, only free the
542 TCPSTAT_INC(tcps_closed);
546 tcp_twrespond(struct tcptw *tw, int flags)
548 struct inpcb *inp = tw->tw_inpcb;
549 #if defined(INET6) || defined(INET)
550 struct tcphdr *th = NULL;
554 struct ip *ip = NULL;
556 u_int hdrlen, optlen;
557 int error = 0; /* Keep compiler happy */
560 struct ip6_hdr *ip6 = NULL;
561 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
563 hdrlen = 0; /* Keep compiler happy */
565 INP_WLOCK_ASSERT(inp);
567 m = m_gethdr(M_NOWAIT, MT_DATA);
570 m->m_data += max_linkhdr;
573 mac_inpcb_create_mbuf(inp, m);
578 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
579 ip6 = mtod(m, struct ip6_hdr *);
580 th = (struct tcphdr *)(ip6 + 1);
581 tcpip_fillheaders(inp, ip6, th);
584 #if defined(INET6) && defined(INET)
589 hdrlen = sizeof(struct tcpiphdr);
590 ip = mtod(m, struct ip *);
591 th = (struct tcphdr *)(ip + 1);
592 tcpip_fillheaders(inp, ip, th);
598 * Send a timestamp and echo-reply if both our side and our peer
599 * have sent timestamps in our SYN's and this is not a RST.
601 if (tw->t_recent && flags == TH_ACK) {
602 to.to_flags |= TOF_TS;
603 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
604 to.to_tsecr = tw->t_recent;
606 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
608 m->m_len = hdrlen + optlen;
609 m->m_pkthdr.len = m->m_len;
611 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
613 th->th_seq = htonl(tw->snd_nxt);
614 th->th_ack = htonl(tw->rcv_nxt);
615 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
616 th->th_flags = flags;
617 th->th_win = htons(tw->last_win);
619 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
622 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
623 th->th_sum = in6_cksum_pseudo(ip6,
624 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
625 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
626 error = ip6_output(m, inp->in6p_outputopts, NULL,
627 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
630 #if defined(INET6) && defined(INET)
635 m->m_pkthdr.csum_flags = CSUM_TCP;
636 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
637 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
638 ip->ip_len = htons(m->m_pkthdr.len);
639 if (V_path_mtu_discovery)
640 ip->ip_off |= htons(IP_DF);
641 error = ip_output(m, inp->inp_options, NULL,
642 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
647 TCPSTAT_INC(tcps_sndacks);
649 TCPSTAT_INC(tcps_sndctrl);
650 TCPSTAT_INC(tcps_sndtotal);
655 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
658 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
659 INP_WLOCK_ASSERT(tw->tw_inpcb);
663 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
664 tw->tw_time = ticks + 2 * tcp_msl;
665 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
666 TW_WUNLOCK(V_tw_lock);
670 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
676 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
682 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
685 TW_WUNLOCK(V_tw_lock);
690 released = in_pcbrele_wlocked(inp);
691 KASSERT(!released, ("%s: inp should not be released here", __func__));
694 uma_zfree(V_tcptw_zone, tw);
698 tcp_tw_2msl_scan(int reuse)
706 * pcbinfo lock is needed in reuse case to prevent deadlock
707 * as two inpcb locks can be acquired simultaneously:
708 * - the inpcb transitioning to TIME_WAIT state in
710 * - the inpcb closed by tcp_twclose().
712 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
718 tw = TAILQ_FIRST(&V_twq_2msl);
719 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
720 TW_RUNLOCK(V_tw_lock);
723 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
728 TW_RUNLOCK(V_tw_lock);
730 if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) {
734 if (in_pcbrele_wlocked(inp)) {
735 KASSERT(tw == NULL, ("%s: held last inp "
736 "reference but tw not NULL", __func__));
737 INP_INFO_WUNLOCK(&V_tcbinfo);
742 /* tcp_twclose() has already been called */
744 INP_INFO_WUNLOCK(&V_tcbinfo);
748 tcp_twclose(tw, reuse);
749 INP_INFO_WUNLOCK(&V_tcbinfo);
753 /* INP_INFO lock is busy, continue later. */
755 if (!in_pcbrele_wlocked(inp))