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_systm.h>
61 #include <netinet/ip.h>
63 #include <netinet/ip6.h>
65 #include <netinet/in_pcb.h>
67 #include <netinet6/in6_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
76 #include <netinet/ip_icmp.h>
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_fsm.h>
79 #include <netinet/tcp_seq.h>
80 #include <netinet/tcp_timer.h>
81 #include <netinet/tcp_var.h>
83 #include <netinet6/tcp6_var.h>
85 #include <netinet/tcpip.h>
87 #include <netinet/tcp_debug.h>
89 #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 * tcbinfo 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 static void tcp_tw_2msl_reset(struct tcptw *, int);
109 static void tcp_tw_2msl_stop(struct tcptw *);
112 tcptw_auto_size(void)
117 * Max out at half the ephemeral port range so that TIME_WAIT
118 * sockets don't tie up too many ephemeral ports.
120 if (V_ipport_lastauto > V_ipport_firstauto)
121 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
123 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
124 /* Protect against goofy port ranges smaller than 32. */
125 return (imin(imax(halfrange, 32), maxsockets / 5));
129 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
134 new = tcptw_auto_size();
137 error = sysctl_handle_int(oidp, &new, 0, req);
138 if (error == 0 && req->newptr)
141 uma_zone_set_max(V_tcptw_zone, maxtcptw);
146 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
147 &maxtcptw, 0, sysctl_maxtcptw, "IU",
148 "Maximum number of compressed TCP TIME_WAIT entries");
150 VNET_DEFINE(int, nolocaltimewait) = 0;
151 #define V_nolocaltimewait VNET(nolocaltimewait)
152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
153 &VNET_NAME(nolocaltimewait), 0,
154 "Do not create compressed TCP TIME_WAIT entries for local connections");
157 tcp_tw_zone_change(void)
161 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
168 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
169 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
170 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
172 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
174 uma_zone_set_max(V_tcptw_zone, maxtcptw);
175 TAILQ_INIT(&V_twq_2msl);
184 INP_INFO_WLOCK(&V_tcbinfo);
185 while((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
187 INP_INFO_WUNLOCK(&V_tcbinfo);
189 uma_zdestroy(V_tcptw_zone);
194 * Move a TCP connection into TIME_WAIT state.
196 * inp is locked, and is unlocked before returning.
199 tcp_twstart(struct tcpcb *tp)
202 struct inpcb *inp = tp->t_inpcb;
206 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */
207 INP_WLOCK_ASSERT(inp);
209 if (V_nolocaltimewait && in_localip(inp->inp_faddr)) {
216 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
218 tw = tcp_tw_2msl_scan(1);
229 * Recover last window size sent.
231 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
234 * Set t_recent if timestamps are used on the connection.
236 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
237 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
238 tw->t_recent = tp->ts_recent;
239 tw->ts_offset = tp->ts_offset;
245 tw->snd_nxt = tp->snd_nxt;
246 tw->rcv_nxt = tp->rcv_nxt;
249 tw->t_starttime = tp->t_starttime;
254 * be used for fin-wait-2 state also, then we may need
255 * a ts_recent from the last segment.
257 acknow = tp->t_flags & TF_ACKNOW;
260 * First, discard tcpcb state, which includes stopping its timers and
261 * freeing it. tcp_discardcb() used to also release the inpcb, but
262 * that work is now done in the caller.
264 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
265 * and might not be needed here any longer.
268 so = inp->inp_socket;
269 soisdisconnected(so);
270 tw->tw_cred = crhold(so->so_cred);
272 tw->tw_so_options = so->so_options;
275 tcp_twrespond(tw, TH_ACK);
277 inp->inp_flags |= INP_TIMEWAIT;
278 tcp_tw_2msl_reset(tw, 0);
281 * If the inpcb owns the sole reference to the socket, then we can
282 * detach and free the socket as it is not needed in time wait.
284 if (inp->inp_flags & INP_SOCKREF) {
285 KASSERT(so->so_state & SS_PROTOREF,
286 ("tcp_twstart: !SS_PROTOREF"));
287 inp->inp_flags &= ~INP_SOCKREF;
291 so->so_state &= ~SS_PROTOREF;
299 * The appromixate rate of ISN increase of Microsoft TCP stacks;
300 * the actual rate is slightly higher due to the addition of
301 * random positive increments.
303 * Most other new OSes use semi-randomized ISN values, so we
304 * do not need to worry about them.
306 #define MS_ISN_BYTES_PER_SECOND 250000
309 * Determine if the ISN we will generate has advanced beyond the last
310 * sequence number used by the previous connection. If so, indicate
311 * that it is safe to recycle this tw socket by returning 1.
314 tcp_twrecycleable(struct tcptw *tw)
316 tcp_seq new_iss = tw->iss;
317 tcp_seq new_irs = tw->irs;
319 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
320 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
321 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
323 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
331 * Returns 1 if the TIME_WAIT state was killed and we should start over,
332 * looking for a pcb in the listen state. Returns 0 otherwise.
335 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
336 struct mbuf *m, int tlen)
342 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */
343 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
344 INP_WLOCK_ASSERT(inp);
347 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
348 * still present. This is undesirable, but temporarily necessary
349 * until we work out how to handle inpcb's who's timewait state has
356 thflags = th->th_flags;
359 * NOTE: for FIN_WAIT_2 (to be added later),
360 * must validate sequence number before accepting RST
364 * If the segment contains RST:
365 * Drop the segment - see Stevens, vol. 2, p. 964 and
368 if (thflags & TH_RST)
372 /* PAWS not needed at the moment */
374 * RFC 1323 PAWS: If we have a timestamp reply on this segment
375 * and it's less than ts_recent, drop it.
377 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
378 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
379 if ((thflags & TH_ACK) == 0)
384 * ts_recent is never updated because we never accept new segments.
389 * If a new connection request is received
390 * while in TIME_WAIT, drop the old connection
391 * and start over if the sequence numbers
392 * are above the previous ones.
394 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
400 * Drop the segment if it does not contain an ACK.
402 if ((thflags & TH_ACK) == 0)
406 * Reset the 2MSL timer if this is a duplicate FIN.
408 if (thflags & TH_FIN) {
409 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
410 if (seq + 1 == tw->rcv_nxt)
411 tcp_tw_2msl_reset(tw, 1);
415 * Acknowledge the segment if it has data or is not a duplicate ACK.
417 if (thflags != TH_ACK || tlen != 0 ||
418 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
419 tcp_twrespond(tw, TH_ACK);
427 tcp_twclose(struct tcptw *tw, int reuse)
433 * At this point, we are in one of two situations:
435 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
438 * (2) We have a socket -- if we own a reference, release it and
439 * notify the socket layer.
442 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
443 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
444 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */
445 INP_WLOCK_ASSERT(inp);
448 tcp_tw_2msl_stop(tw);
449 inp->inp_ppcb = NULL;
452 so = inp->inp_socket;
455 * If there's a socket, handle two cases: first, we own a
456 * strong reference, which we will now release, or we don't
457 * in which case another reference exists (XXXRW: think
458 * about this more), and we don't need to take action.
460 if (inp->inp_flags & INP_SOCKREF) {
461 inp->inp_flags &= ~INP_SOCKREF;
465 KASSERT(so->so_state & SS_PROTOREF,
466 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
467 so->so_state &= ~SS_PROTOREF;
471 * If we don't own the only reference, the socket and
472 * inpcb need to be left around to be handled by
473 * tcp_usr_detach() later.
479 TCPSTAT_INC(tcps_closed);
484 uma_zfree(V_tcptw_zone, tw);
488 tcp_twrespond(struct tcptw *tw, int flags)
490 struct inpcb *inp = tw->tw_inpcb;
493 struct ip *ip = NULL;
494 u_int hdrlen, optlen;
498 struct ip6_hdr *ip6 = NULL;
499 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
502 INP_WLOCK_ASSERT(inp);
504 m = m_gethdr(M_DONTWAIT, MT_DATA);
507 m->m_data += max_linkhdr;
510 mac_inpcb_create_mbuf(inp, m);
515 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
516 ip6 = mtod(m, struct ip6_hdr *);
517 th = (struct tcphdr *)(ip6 + 1);
518 tcpip_fillheaders(inp, ip6, th);
522 hdrlen = sizeof(struct tcpiphdr);
523 ip = mtod(m, struct ip *);
524 th = (struct tcphdr *)(ip + 1);
525 tcpip_fillheaders(inp, ip, th);
530 * Send a timestamp and echo-reply if both our side and our peer
531 * have sent timestamps in our SYN's and this is not a RST.
533 if (tw->t_recent && flags == TH_ACK) {
534 to.to_flags |= TOF_TS;
535 to.to_tsval = ticks + tw->ts_offset;
536 to.to_tsecr = tw->t_recent;
538 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
540 m->m_len = hdrlen + optlen;
541 m->m_pkthdr.len = m->m_len;
543 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
545 th->th_seq = htonl(tw->snd_nxt);
546 th->th_ack = htonl(tw->rcv_nxt);
547 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
548 th->th_flags = flags;
549 th->th_win = htons(tw->last_win);
553 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
554 sizeof(struct tcphdr) + optlen);
555 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
556 error = ip6_output(m, inp->in6p_outputopts, NULL,
557 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
561 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
562 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
563 m->m_pkthdr.csum_flags = CSUM_TCP;
564 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
565 ip->ip_len = m->m_pkthdr.len;
566 if (V_path_mtu_discovery)
568 error = ip_output(m, inp->inp_options, NULL,
569 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
573 TCPSTAT_INC(tcps_sndacks);
575 TCPSTAT_INC(tcps_sndctrl);
576 TCPSTAT_INC(tcps_sndtotal);
581 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
584 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
585 INP_WLOCK_ASSERT(tw->tw_inpcb);
587 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
588 tw->tw_time = ticks + 2 * tcp_msl;
589 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
593 tcp_tw_2msl_stop(struct tcptw *tw)
596 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
597 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
601 tcp_tw_2msl_scan(int reuse)
605 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
607 tw = TAILQ_FIRST(&V_twq_2msl);
608 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0))
610 INP_WLOCK(tw->tw_inpcb);
611 tcp_twclose(tw, reuse);