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
34 #include "opt_inet6.h"
36 #include "opt_tcpdebug.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/malloc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/random.h>
54 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
61 #include <netinet/ip6.h>
63 #include <netinet/in_pcb.h>
65 #include <netinet6/in6_pcb.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip_var.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
74 #include <netinet/ip_icmp.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>
87 #include <netinet6/ip6protosw.h>
89 #include <machine/in_cksum.h>
91 #include <security/mac/mac_framework.h>
93 static uma_zone_t tcptw_zone;
97 * The timed wait queue contains references to each of the TCP sessions
98 * currently in the TIME_WAIT state. The queue pointers, including the
99 * queue pointers in each tcptw structure, are protected using the global
100 * tcbinfo lock, which must be held over queue iteration and modification.
102 static TAILQ_HEAD(, tcptw) twq_2msl;
104 static void tcp_tw_2msl_reset(struct tcptw *, int);
105 static void tcp_tw_2msl_stop(struct tcptw *);
108 tcptw_auto_size(void)
113 * Max out at half the ephemeral port range so that TIME_WAIT
114 * sockets don't tie up too many ephemeral ports.
116 if (ipport_lastauto > ipport_firstauto)
117 halfrange = (ipport_lastauto - ipport_firstauto) / 2;
119 halfrange = (ipport_firstauto - ipport_lastauto) / 2;
120 /* Protect against goofy port ranges smaller than 32. */
121 return (imin(imax(halfrange, 32), maxsockets / 5));
125 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
130 new = tcptw_auto_size();
133 error = sysctl_handle_int(oidp, &new, 0, req);
134 if (error == 0 && req->newptr)
137 uma_zone_set_max(tcptw_zone, maxtcptw);
142 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
143 &maxtcptw, 0, sysctl_maxtcptw, "IU",
144 "Maximum number of compressed TCP TIME_WAIT entries");
146 static int nolocaltimewait = 0;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
149 "Do not create compressed TCP TIME_WAIT entries for local connections");
152 tcp_tw_zone_change(void)
156 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
163 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
164 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
165 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
167 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
169 uma_zone_set_max(tcptw_zone, maxtcptw);
170 TAILQ_INIT(&twq_2msl);
174 * Move a TCP connection into TIME_WAIT state.
176 * inp is locked, and is unlocked before returning.
179 tcp_twstart(struct tcpcb *tp)
182 struct inpcb *inp = tp->t_inpcb;
186 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_tw_2msl_reset(). */
187 INP_LOCK_ASSERT(inp);
189 if (nolocaltimewait && in_localip(inp->inp_faddr)) {
196 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
198 tw = tcp_tw_2msl_scan(1);
209 * Recover last window size sent.
211 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
214 * Set t_recent if timestamps are used on the connection.
216 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
217 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
218 tw->t_recent = tp->ts_recent;
219 tw->ts_offset = tp->ts_offset;
225 tw->snd_nxt = tp->snd_nxt;
226 tw->rcv_nxt = tp->rcv_nxt;
229 tw->t_starttime = tp->t_starttime;
234 * be used for fin-wait-2 state also, then we may need
235 * a ts_recent from the last segment.
237 acknow = tp->t_flags & TF_ACKNOW;
240 * First, discard tcpcb state, which includes stopping its timers and
241 * freeing it. tcp_discardcb() used to also release the inpcb, but
242 * that work is now done in the caller.
244 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
245 * and might not be needed here any longer.
248 so = inp->inp_socket;
249 soisdisconnected(so);
250 tw->tw_cred = crhold(so->so_cred);
252 tw->tw_so_options = so->so_options;
255 tcp_twrespond(tw, TH_ACK);
257 inp->inp_vflag |= INP_TIMEWAIT;
258 tcp_tw_2msl_reset(tw, 0);
261 * If the inpcb owns the sole reference to the socket, then we can
262 * detach and free the socket as it is not needed in time wait.
264 if (inp->inp_vflag & INP_SOCKREF) {
265 KASSERT(so->so_state & SS_PROTOREF,
266 ("tcp_twstart: !SS_PROTOREF"));
267 inp->inp_vflag &= ~INP_SOCKREF;
271 so->so_state &= ~SS_PROTOREF;
279 * The appromixate rate of ISN increase of Microsoft TCP stacks;
280 * the actual rate is slightly higher due to the addition of
281 * random positive increments.
283 * Most other new OSes use semi-randomized ISN values, so we
284 * do not need to worry about them.
286 #define MS_ISN_BYTES_PER_SECOND 250000
289 * Determine if the ISN we will generate has advanced beyond the last
290 * sequence number used by the previous connection. If so, indicate
291 * that it is safe to recycle this tw socket by returning 1.
294 tcp_twrecycleable(struct tcptw *tw)
296 tcp_seq new_iss = tw->iss;
297 tcp_seq new_irs = tw->irs;
299 INP_INFO_WLOCK_ASSERT(&tcbinfo);
300 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
301 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
303 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
311 * Returns 1 if the TIME_WAIT state was killed and we should start over,
312 * looking for a pcb in the listen state. Returns 0 otherwise.
315 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
316 struct mbuf *m, int tlen)
322 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
324 const int isipv6 = 0;
327 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */
328 INP_INFO_WLOCK_ASSERT(&tcbinfo);
329 INP_LOCK_ASSERT(inp);
332 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
333 * still present. This is undesirable, but temporarily necessary
334 * until we work out how to handle inpcb's who's timewait state has
341 thflags = th->th_flags;
344 * NOTE: for FIN_WAIT_2 (to be added later),
345 * must validate sequence number before accepting RST
349 * If the segment contains RST:
350 * Drop the segment - see Stevens, vol. 2, p. 964 and
353 if (thflags & TH_RST)
357 /* PAWS not needed at the moment */
359 * RFC 1323 PAWS: If we have a timestamp reply on this segment
360 * and it's less than ts_recent, drop it.
362 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
363 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
364 if ((thflags & TH_ACK) == 0)
369 * ts_recent is never updated because we never accept new segments.
374 * If a new connection request is received
375 * while in TIME_WAIT, drop the old connection
376 * and start over if the sequence numbers
377 * are above the previous ones.
379 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
385 * Drop the the segment if it does not contain an ACK.
387 if ((thflags & TH_ACK) == 0)
391 * Reset the 2MSL timer if this is a duplicate FIN.
393 if (thflags & TH_FIN) {
394 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
395 if (seq + 1 == tw->rcv_nxt)
396 tcp_tw_2msl_reset(tw, 1);
400 * Acknowledge the segment if it has data or is not a duplicate ACK.
402 if (thflags != TH_ACK || tlen != 0 ||
403 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
404 tcp_twrespond(tw, TH_ACK);
408 * Generate a RST, dropping incoming segment.
409 * Make ACK acceptable to originator of segment.
410 * Don't bother to respond if destination was broadcast/multicast.
412 if (m->m_flags & (M_BCAST|M_MCAST))
418 /* IPv6 anycast check is done at tcp6_input() */
419 ip6 = mtod(m, struct ip6_hdr *);
420 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
421 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
427 ip = mtod(m, struct ip *);
428 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
429 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
430 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
431 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
434 if (thflags & TH_ACK) {
436 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
438 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
440 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
452 tcp_twclose(struct tcptw *tw, int reuse)
458 * At this point, we are in one of two situations:
460 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
463 * (2) We have a socket -- if we own a reference, release it and
464 * notify the socket layer.
467 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
468 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
469 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_tw_2msl_stop(). */
470 INP_LOCK_ASSERT(inp);
473 tcp_tw_2msl_stop(tw);
474 inp->inp_ppcb = NULL;
477 so = inp->inp_socket;
480 * If there's a socket, handle two cases: first, we own a
481 * strong reference, which we will now release, or we don't
482 * in which case another reference exists (XXXRW: think
483 * about this more), and we don't need to take action.
485 if (inp->inp_vflag & INP_SOCKREF) {
486 inp->inp_vflag &= ~INP_SOCKREF;
490 KASSERT(so->so_state & SS_PROTOREF,
491 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
492 so->so_state &= ~SS_PROTOREF;
496 * If we don't own the only reference, the socket and
497 * inpcb need to be left around to be handled by
498 * tcp_usr_detach() later.
504 if (inp->inp_vflag & INP_IPV6PROTO)
510 tcpstat.tcps_closed++;
515 uma_zfree(tcptw_zone, tw);
519 tcp_twrespond(struct tcptw *tw, int flags)
521 struct inpcb *inp = tw->tw_inpcb;
524 struct ip *ip = NULL;
525 u_int hdrlen, optlen;
529 struct ip6_hdr *ip6 = NULL;
530 int isipv6 = inp->inp_inc.inc_isipv6;
533 INP_LOCK_ASSERT(inp);
535 m = m_gethdr(M_DONTWAIT, MT_DATA);
538 m->m_data += max_linkhdr;
541 mac_create_mbuf_from_inpcb(inp, m);
546 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
547 ip6 = mtod(m, struct ip6_hdr *);
548 th = (struct tcphdr *)(ip6 + 1);
549 tcpip_fillheaders(inp, ip6, th);
553 hdrlen = sizeof(struct tcpiphdr);
554 ip = mtod(m, struct ip *);
555 th = (struct tcphdr *)(ip + 1);
556 tcpip_fillheaders(inp, ip, th);
561 * Send a timestamp and echo-reply if both our side and our peer
562 * have sent timestamps in our SYN's and this is not a RST.
564 if (tw->t_recent && flags == TH_ACK) {
565 to.to_flags |= TOF_TS;
566 to.to_tsval = ticks + tw->ts_offset;
567 to.to_tsecr = tw->t_recent;
569 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
571 m->m_len = hdrlen + optlen;
572 m->m_pkthdr.len = m->m_len;
574 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
576 th->th_seq = htonl(tw->snd_nxt);
577 th->th_ack = htonl(tw->rcv_nxt);
578 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
579 th->th_flags = flags;
580 th->th_win = htons(tw->last_win);
584 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
585 sizeof(struct tcphdr) + optlen);
586 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
587 error = ip6_output(m, inp->in6p_outputopts, NULL,
588 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
592 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
593 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
594 m->m_pkthdr.csum_flags = CSUM_TCP;
595 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
596 ip->ip_len = m->m_pkthdr.len;
597 if (path_mtu_discovery)
599 error = ip_output(m, inp->inp_options, NULL,
600 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
604 tcpstat.tcps_sndacks++;
606 tcpstat.tcps_sndctrl++;
607 tcpstat.tcps_sndtotal++;
612 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
615 INP_INFO_WLOCK_ASSERT(&tcbinfo);
616 INP_LOCK_ASSERT(tw->tw_inpcb);
618 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
619 tw->tw_time = ticks + 2 * tcp_msl;
620 TAILQ_INSERT_TAIL(&twq_2msl, tw, tw_2msl);
624 tcp_tw_2msl_stop(struct tcptw *tw)
627 INP_INFO_WLOCK_ASSERT(&tcbinfo);
628 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
632 tcp_tw_2msl_scan(int reuse)
636 INP_INFO_WLOCK_ASSERT(&tcbinfo);
638 tw = TAILQ_FIRST(&twq_2msl);
639 if (tw == NULL || (!reuse && tw->tw_time > ticks))
641 INP_LOCK(tw->tw_inpcb);
642 tcp_twclose(tw, reuse);