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>
52 #include <sys/refcount.h>
56 #include <net/route.h>
58 #include <net/if_var.h>
61 #include <netinet/in.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_icmp.h>
67 #include <netinet/ip_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_pcb.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/scope6_var.h>
73 #include <netinet6/nd6.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_fsm.h>
77 #include <netinet/tcp_seq.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
83 #include <netinet/tcpip.h>
85 #include <netinet/tcp_debug.h>
88 #include <netinet6/ip6protosw.h>
91 #include <machine/in_cksum.h>
93 #include <security/mac/mac_framework.h>
95 static VNET_DEFINE(uma_zone_t, tcptw_zone);
96 #define V_tcptw_zone VNET(tcptw_zone)
100 * The timed wait queue contains references to each of the TCP sessions
101 * currently in the TIME_WAIT state. The queue pointers, including the
102 * queue pointers in each tcptw structure, are protected using the global
103 * timewait lock, which must be held over queue iteration and modification.
105 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
106 #define V_twq_2msl VNET(twq_2msl)
108 /* Global timewait lock */
109 static VNET_DEFINE(struct rwlock, tw_lock);
110 #define V_tw_lock VNET(tw_lock)
112 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0)
113 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw))
114 #define TW_RLOCK(tw) rw_rlock(&(tw))
115 #define TW_WLOCK(tw) rw_wlock(&(tw))
116 #define TW_RUNLOCK(tw) rw_runlock(&(tw))
117 #define TW_WUNLOCK(tw) rw_wunlock(&(tw))
118 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED)
119 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED)
120 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED)
121 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED)
123 static void tcp_tw_2msl_reset(struct tcptw *, int);
124 static void tcp_tw_2msl_stop(struct tcptw *, int);
127 * tw_pcbref() bumps the reference count on an tw in order to maintain
128 * stability of an tw pointer despite the tw lock being released.
131 tw_pcbref(struct tcptw *tw)
134 KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
135 refcount_acquire(&tw->tw_refcount);
139 * Drop a refcount on an tw elevated using tw_pcbref().
142 tw_pcbrele(struct tcptw *tw)
145 KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
146 if (!refcount_release(&tw->tw_refcount))
148 uma_zfree(V_tcptw_zone, tw);
153 tcptw_auto_size(void)
158 * Max out at half the ephemeral port range so that TIME_WAIT
159 * sockets don't tie up too many ephemeral ports.
161 if (V_ipport_lastauto > V_ipport_firstauto)
162 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
164 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
165 /* Protect against goofy port ranges smaller than 32. */
166 return (imin(imax(halfrange, 32), maxsockets / 5));
170 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
175 new = tcptw_auto_size();
178 error = sysctl_handle_int(oidp, &new, 0, req);
179 if (error == 0 && req->newptr)
182 uma_zone_set_max(V_tcptw_zone, maxtcptw);
187 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
188 &maxtcptw, 0, sysctl_maxtcptw, "IU",
189 "Maximum number of compressed TCP TIME_WAIT entries");
191 VNET_DEFINE(int, nolocaltimewait) = 0;
192 #define V_nolocaltimewait VNET(nolocaltimewait)
193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
194 &VNET_NAME(nolocaltimewait), 0,
195 "Do not create compressed TCP TIME_WAIT entries for local connections");
198 tcp_tw_zone_change(void)
202 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
209 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
210 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
211 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
213 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
215 uma_zone_set_max(V_tcptw_zone, maxtcptw);
216 TAILQ_INIT(&V_twq_2msl);
217 TW_LOCK_INIT(V_tw_lock, "tcptw");
226 INP_INFO_WLOCK(&V_tcbinfo);
227 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
229 INP_INFO_WUNLOCK(&V_tcbinfo);
231 TW_LOCK_DESTROY(V_tw_lock);
232 uma_zdestroy(V_tcptw_zone);
237 * Move a TCP connection into TIME_WAIT state.
239 * inp is locked, and is unlocked before returning.
242 tcp_twstart(struct tcpcb *tp)
245 struct inpcb *inp = tp->t_inpcb;
249 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
252 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
253 INP_WLOCK_ASSERT(inp);
255 if (V_nolocaltimewait) {
259 error = in6_localaddr(&inp->in6p_faddr);
261 #if defined(INET6) && defined(INET)
265 error = in_localip(inp->inp_faddr);
275 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
277 tw = tcp_tw_2msl_reuse();
286 refcount_init(&tw->tw_refcount, 1);
289 * Recover last window size sent.
291 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
292 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
297 * Set t_recent if timestamps are used on the connection.
299 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
300 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
301 tw->t_recent = tp->ts_recent;
302 tw->ts_offset = tp->ts_offset;
308 tw->snd_nxt = tp->snd_nxt;
309 tw->rcv_nxt = tp->rcv_nxt;
312 tw->t_starttime = tp->t_starttime;
317 * be used for fin-wait-2 state also, then we may need
318 * a ts_recent from the last segment.
320 acknow = tp->t_flags & TF_ACKNOW;
323 * First, discard tcpcb state, which includes stopping its timers and
324 * freeing it. tcp_discardcb() used to also release the inpcb, but
325 * that work is now done in the caller.
327 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
328 * and might not be needed here any longer.
331 so = inp->inp_socket;
332 soisdisconnected(so);
333 tw->tw_cred = crhold(so->so_cred);
335 tw->tw_so_options = so->so_options;
338 tcp_twrespond(tw, TH_ACK);
340 inp->inp_flags |= INP_TIMEWAIT;
341 tcp_tw_2msl_reset(tw, 0);
344 * If the inpcb owns the sole reference to the socket, then we can
345 * detach and free the socket as it is not needed in time wait.
347 if (inp->inp_flags & INP_SOCKREF) {
348 KASSERT(so->so_state & SS_PROTOREF,
349 ("tcp_twstart: !SS_PROTOREF"));
350 inp->inp_flags &= ~INP_SOCKREF;
354 so->so_state &= ~SS_PROTOREF;
361 * Returns 1 if the TIME_WAIT state was killed and we should start over,
362 * looking for a pcb in the listen state. Returns 0 otherwise.
365 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
366 struct mbuf *m, int tlen)
372 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
373 INP_WLOCK_ASSERT(inp);
376 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
377 * still present. This is undesirable, but temporarily necessary
378 * until we work out how to handle inpcb's who's timewait state has
385 thflags = th->th_flags;
388 * NOTE: for FIN_WAIT_2 (to be added later),
389 * must validate sequence number before accepting RST
393 * If the segment contains RST:
394 * Drop the segment - see Stevens, vol. 2, p. 964 and
397 if (thflags & TH_RST)
401 /* PAWS not needed at the moment */
403 * RFC 1323 PAWS: If we have a timestamp reply on this segment
404 * and it's less than ts_recent, drop it.
406 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
407 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
408 if ((thflags & TH_ACK) == 0)
413 * ts_recent is never updated because we never accept new segments.
418 * If a new connection request is received
419 * while in TIME_WAIT, drop the old connection
420 * and start over if the sequence numbers
421 * are above the previous ones.
423 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
429 * Drop the segment if it does not contain an ACK.
431 if ((thflags & TH_ACK) == 0)
435 * Reset the 2MSL timer if this is a duplicate FIN.
437 if (thflags & TH_FIN) {
438 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
439 if (seq + 1 == tw->rcv_nxt)
440 tcp_tw_2msl_reset(tw, 1);
444 * Acknowledge the segment if it has data or is not a duplicate ACK.
446 if (thflags != TH_ACK || tlen != 0 ||
447 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
448 tcp_twrespond(tw, TH_ACK);
456 tcp_twclose(struct tcptw *tw, int reuse)
462 * At this point, we are in one of two situations:
464 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
467 * (2) We have a socket -- if we own a reference, release it and
468 * notify the socket layer.
471 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
472 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
473 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */
474 INP_WLOCK_ASSERT(inp);
477 tcp_tw_2msl_stop(tw, reuse);
478 inp->inp_ppcb = NULL;
481 so = inp->inp_socket;
484 * If there's a socket, handle two cases: first, we own a
485 * strong reference, which we will now release, or we don't
486 * in which case another reference exists (XXXRW: think
487 * about this more), and we don't need to take action.
489 if (inp->inp_flags & INP_SOCKREF) {
490 inp->inp_flags &= ~INP_SOCKREF;
494 KASSERT(so->so_state & SS_PROTOREF,
495 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
496 so->so_state &= ~SS_PROTOREF;
500 * If we don't own the only reference, the socket and
501 * inpcb need to be left around to be handled by
502 * tcp_usr_detach() later.
508 TCPSTAT_INC(tcps_closed);
512 tcp_twrespond(struct tcptw *tw, int flags)
514 struct inpcb *inp = tw->tw_inpcb;
515 #if defined(INET6) || defined(INET)
516 struct tcphdr *th = NULL;
520 struct ip *ip = NULL;
522 u_int hdrlen, optlen;
523 int error = 0; /* Keep compiler happy */
526 struct ip6_hdr *ip6 = NULL;
527 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
529 hdrlen = 0; /* Keep compiler happy */
531 INP_WLOCK_ASSERT(inp);
533 m = m_gethdr(M_NOWAIT, MT_DATA);
536 m->m_data += max_linkhdr;
539 mac_inpcb_create_mbuf(inp, m);
544 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
545 ip6 = mtod(m, struct ip6_hdr *);
546 th = (struct tcphdr *)(ip6 + 1);
547 tcpip_fillheaders(inp, ip6, th);
550 #if defined(INET6) && defined(INET)
555 hdrlen = sizeof(struct tcpiphdr);
556 ip = mtod(m, struct ip *);
557 th = (struct tcphdr *)(ip + 1);
558 tcpip_fillheaders(inp, ip, th);
564 * Send a timestamp and echo-reply if both our side and our peer
565 * have sent timestamps in our SYN's and this is not a RST.
567 if (tw->t_recent && flags == TH_ACK) {
568 to.to_flags |= TOF_TS;
569 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
570 to.to_tsecr = tw->t_recent;
572 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
574 m->m_len = hdrlen + optlen;
575 m->m_pkthdr.len = m->m_len;
577 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
579 th->th_seq = htonl(tw->snd_nxt);
580 th->th_ack = htonl(tw->rcv_nxt);
581 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
582 th->th_flags = flags;
583 th->th_win = htons(tw->last_win);
585 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
588 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
589 th->th_sum = in6_cksum_pseudo(ip6,
590 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
591 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
592 error = ip6_output(m, inp->in6p_outputopts, NULL,
593 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
596 #if defined(INET6) && defined(INET)
601 m->m_pkthdr.csum_flags = CSUM_TCP;
602 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
603 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
604 ip->ip_len = htons(m->m_pkthdr.len);
605 if (V_path_mtu_discovery)
606 ip->ip_off |= htons(IP_DF);
607 error = ip_output(m, inp->inp_options, NULL,
608 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
613 TCPSTAT_INC(tcps_sndacks);
615 TCPSTAT_INC(tcps_sndctrl);
616 TCPSTAT_INC(tcps_sndtotal);
621 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
624 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
625 INP_WLOCK_ASSERT(tw->tw_inpcb);
629 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
630 tw->tw_time = ticks + 2 * tcp_msl;
631 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
632 TW_WUNLOCK(V_tw_lock);
636 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
639 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
642 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
645 TW_WUNLOCK(V_tw_lock);
652 tcp_tw_2msl_reuse(void)
656 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
659 tw = TAILQ_FIRST(&V_twq_2msl);
661 TW_WUNLOCK(V_tw_lock);
664 TW_WUNLOCK(V_tw_lock);
666 INP_WLOCK(tw->tw_inpcb);
673 tcp_tw_2msl_scan(void)
679 tw = TAILQ_FIRST(&V_twq_2msl);
680 if (tw == NULL || tw->tw_time - ticks > 0) {
681 TW_RUNLOCK(V_tw_lock);
685 TW_RUNLOCK(V_tw_lock);
687 /* Close timewait state */
688 if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) {
689 if (tw_pcbrele(tw)) {
690 INP_INFO_WUNLOCK(&V_tcbinfo);
694 KASSERT(tw->tw_inpcb != NULL,
695 ("%s: tw->tw_inpcb == NULL", __func__));
696 INP_WLOCK(tw->tw_inpcb);
698 INP_INFO_WUNLOCK(&V_tcbinfo);
700 /* INP_INFO lock is busy; continue later. */