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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
50 #include <sys/domain.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
60 #include <net/route.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
70 #include <netinet/in_pcb.h>
72 #include <netinet6/in6_pcb.h>
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
77 #include <netinet6/ip6_var.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>
91 #include <netinet6/ip6protosw.h>
94 #include <netinet6/ipsec.h>
96 #include <netinet6/ipsec6.h>
100 #include <machine/in_cksum.h>
103 int tcp_mssdflt = TCP_MSS;
104 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
105 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
108 int tcp_v6mssdflt = TCP6_MSS;
109 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
110 CTLFLAG_RW, &tcp_v6mssdflt , 0,
111 "Default TCP Maximum Segment Size for IPv6");
115 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
116 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
117 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
120 int tcp_do_rfc1323 = 1;
121 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
122 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
124 int tcp_do_rfc1644 = 0;
125 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
126 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
128 static int tcp_tcbhashsize = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
130 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
132 static int do_tcpdrain = 1;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
134 "Enable tcp_drain routine for extra help when low on mbufs");
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
137 &tcbinfo.ipi_count, 0, "Number of active PCBs");
139 static int icmp_may_rst = 1;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
141 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
143 static int tcp_isn_reseed_interval = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
145 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
147 static void tcp_cleartaocache(void);
148 static struct inpcb *tcp_notify(struct inpcb *, int);
151 * Target size of TCP PCB hash tables. Must be a power of two.
153 * Note that this can be overridden by the kernel environment
154 * variable net.inet.tcp.tcbhashsize
157 #define TCBHASHSIZE 512
161 * This is the actual shape of what we allocate using the zone
162 * allocator. Doing it this way allows us to protect both structures
163 * using the same generation count, and also eliminates the overhead
164 * of allocating tcpcbs separately. By hiding the structure here,
165 * we avoid changing most of the rest of the code (although it needs
166 * to be changed, eventually, for greater efficiency).
169 #define ALIGNM1 (ALIGNMENT - 1)
173 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
176 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
177 struct callout inp_tp_delack;
188 int hashsize = TCBHASHSIZE;
193 tcp_delacktime = TCPTV_DELACK;
194 tcp_keepinit = TCPTV_KEEP_INIT;
195 tcp_keepidle = TCPTV_KEEP_IDLE;
196 tcp_keepintvl = TCPTV_KEEPINTVL;
197 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
199 tcp_rexmit_min = TCPTV_MIN;
200 tcp_rexmit_slop = TCPTV_CPU_VAR;
202 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
204 tcbinfo.listhead = &tcb;
205 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
206 if (!powerof2(hashsize)) {
207 printf("WARNING: TCB hash size not a power of 2\n");
208 hashsize = 512; /* safe default */
210 tcp_tcbhashsize = hashsize;
211 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
212 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
213 &tcbinfo.porthashmask);
214 tcbinfo.ipi_zone = uma_zcreate("tcpcb", sizeof(struct inp_tp),
215 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
216 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
218 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
220 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
222 if (max_protohdr < TCP_MINPROTOHDR)
223 max_protohdr = TCP_MINPROTOHDR;
224 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
226 #undef TCP_MINPROTOHDR
232 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
233 * tcp_template used to store this data in mbufs, but we now recopy it out
234 * of the tcpcb each time to conserve mbufs.
237 tcp_fillheaders(tp, ip_ptr, tcp_ptr)
242 struct inpcb *inp = tp->t_inpcb;
243 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
246 if ((inp->inp_vflag & INP_IPV6) != 0) {
249 ip6 = (struct ip6_hdr *)ip_ptr;
250 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
251 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
252 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
253 (IPV6_VERSION & IPV6_VERSION_MASK);
254 ip6->ip6_nxt = IPPROTO_TCP;
255 ip6->ip6_plen = sizeof(struct tcphdr);
256 ip6->ip6_src = inp->in6p_laddr;
257 ip6->ip6_dst = inp->in6p_faddr;
262 struct ip *ip = (struct ip *) ip_ptr;
264 ip->ip_vhl = IP_VHL_BORING;
271 ip->ip_p = IPPROTO_TCP;
272 ip->ip_src = inp->inp_laddr;
273 ip->ip_dst = inp->inp_faddr;
274 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
275 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
278 tcp_hdr->th_sport = inp->inp_lport;
279 tcp_hdr->th_dport = inp->inp_fport;
284 tcp_hdr->th_flags = 0;
290 * Create template to be used to send tcp packets on a connection.
291 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
292 * use for this function is in keepalives, which use tcp_respond.
301 m = m_get(M_DONTWAIT, MT_HEADER);
304 m->m_len = sizeof(struct tcptemp);
305 n = mtod(m, struct tcptemp *);
307 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
312 * Send a single message to the TCP at address specified by
313 * the given TCP/IP header. If m == 0, then we make a copy
314 * of the tcpiphdr at ti and send directly to the addressed host.
315 * This is used to force keep alive messages out using the TCP
316 * template for a connection. If flags are given then we send
317 * a message back to the TCP which originated the * segment ti,
318 * and discard the mbuf containing it and any other attached mbufs.
320 * In any case the ack and sequence number of the transmitted
321 * segment are as specified by the parameters.
323 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
326 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
329 register struct tcphdr *th;
330 register struct mbuf *m;
336 struct route *ro = 0;
341 struct route_in6 *ro6 = 0;
342 struct route_in6 sro6;
349 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
355 if (!(flags & TH_RST)) {
356 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
357 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
358 win = (long)TCP_MAXWIN << tp->rcv_scale;
362 ro6 = &tp->t_inpcb->in6p_route;
365 ro = &tp->t_inpcb->inp_route;
370 bzero(ro6, sizeof *ro6);
375 bzero(ro, sizeof *ro);
379 m = m_gethdr(M_DONTWAIT, MT_HEADER);
383 m->m_data += max_linkhdr;
386 bcopy((caddr_t)ip6, mtod(m, caddr_t),
387 sizeof(struct ip6_hdr));
388 ip6 = mtod(m, struct ip6_hdr *);
389 nth = (struct tcphdr *)(ip6 + 1);
393 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
394 ip = mtod(m, struct ip *);
395 nth = (struct tcphdr *)(ip + 1);
397 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
402 m->m_data = (caddr_t)ipgen;
403 /* m_len is set later */
405 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
408 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
409 nth = (struct tcphdr *)(ip6 + 1);
413 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
414 nth = (struct tcphdr *)(ip + 1);
418 * this is usually a case when an extension header
419 * exists between the IPv6 header and the
422 nth->th_sport = th->th_sport;
423 nth->th_dport = th->th_dport;
425 xchg(nth->th_dport, nth->th_sport, n_short);
431 ip6->ip6_vfc = IPV6_VERSION;
432 ip6->ip6_nxt = IPPROTO_TCP;
433 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
435 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
439 tlen += sizeof (struct tcpiphdr);
441 ip->ip_ttl = ip_defttl;
444 m->m_pkthdr.len = tlen;
445 m->m_pkthdr.rcvif = (struct ifnet *) 0;
446 nth->th_seq = htonl(seq);
447 nth->th_ack = htonl(ack);
449 nth->th_off = sizeof (struct tcphdr) >> 2;
450 nth->th_flags = flags;
452 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
454 nth->th_win = htons((u_short)win);
459 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
460 sizeof(struct ip6_hdr),
461 tlen - sizeof(struct ip6_hdr));
462 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
469 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
470 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
471 m->m_pkthdr.csum_flags = CSUM_TCP;
472 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
475 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
476 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
479 if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
486 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
487 if (ro6 == &sro6 && ro6->ro_rt) {
494 (void) ip_output(m, NULL, ro, ipflags, NULL);
495 if (ro == &sro && ro->ro_rt) {
503 * Create a new TCP control block, making an
504 * empty reassembly queue and hooking it to the argument
505 * protocol control block. The `inp' parameter must have
506 * come from the zone allocator set up in tcp_init().
513 register struct tcpcb *tp;
515 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
518 it = (struct inp_tp *)inp;
520 bzero((char *) tp, sizeof(struct tcpcb));
521 LIST_INIT(&tp->t_segq);
522 tp->t_maxseg = tp->t_maxopd =
524 isipv6 ? tcp_v6mssdflt :
528 /* Set up our timeouts. */
529 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0);
530 callout_init(tp->tt_persist = &it->inp_tp_persist, 0);
531 callout_init(tp->tt_keep = &it->inp_tp_keep, 0);
532 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0);
533 callout_init(tp->tt_delack = &it->inp_tp_delack, 0);
536 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
538 tp->t_flags |= TF_REQ_CC;
539 tp->t_inpcb = inp; /* XXX */
541 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
542 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
543 * reasonable initial retransmit time.
545 tp->t_srtt = TCPTV_SRTTBASE;
546 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
547 tp->t_rttmin = tcp_rexmit_min;
548 tp->t_rxtcur = TCPTV_RTOBASE;
549 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
550 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
551 tp->t_rcvtime = ticks;
553 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
554 * because the socket may be bound to an IPv6 wildcard address,
555 * which may match an IPv4-mapped IPv6 address.
557 inp->inp_ip_ttl = ip_defttl;
558 inp->inp_ppcb = (caddr_t)tp;
559 return (tp); /* XXX */
563 * Drop a TCP connection, reporting
564 * the specified error. If connection is synchronized,
565 * then send a RST to peer.
569 register struct tcpcb *tp;
572 struct socket *so = tp->t_inpcb->inp_socket;
574 if (TCPS_HAVERCVDSYN(tp->t_state)) {
575 tp->t_state = TCPS_CLOSED;
576 (void) tcp_output(tp);
577 tcpstat.tcps_drops++;
579 tcpstat.tcps_conndrops++;
580 if (errno == ETIMEDOUT && tp->t_softerror)
581 errno = tp->t_softerror;
582 so->so_error = errno;
583 return (tcp_close(tp));
587 * Close a TCP control block:
588 * discard all space held by the tcp
589 * discard internet protocol block
590 * wake up any sleepers
594 register struct tcpcb *tp;
596 register struct tseg_qent *q;
597 struct inpcb *inp = tp->t_inpcb;
598 struct socket *so = inp->inp_socket;
600 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
602 register struct rtentry *rt;
606 * Make sure that all of our timers are stopped before we
609 callout_stop(tp->tt_rexmt);
610 callout_stop(tp->tt_persist);
611 callout_stop(tp->tt_keep);
612 callout_stop(tp->tt_2msl);
613 callout_stop(tp->tt_delack);
616 * If we got enough samples through the srtt filter,
617 * save the rtt and rttvar in the routing entry.
618 * 'Enough' is arbitrarily defined as the 16 samples.
619 * 16 samples is enough for the srtt filter to converge
620 * to within 5% of the correct value; fewer samples and
621 * we could save a very bogus rtt.
623 * Don't update the default route's characteristics and don't
624 * update anything that the user "locked".
626 if (tp->t_rttupdated >= 16) {
627 register u_long i = 0;
630 struct sockaddr_in6 *sin6;
632 if ((rt = inp->in6p_route.ro_rt) == NULL)
634 sin6 = (struct sockaddr_in6 *)rt_key(rt);
635 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
640 if ((rt = inp->inp_route.ro_rt) == NULL ||
641 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
645 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
647 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
648 if (rt->rt_rmx.rmx_rtt && i)
650 * filter this update to half the old & half
651 * the new values, converting scale.
652 * See route.h and tcp_var.h for a
653 * description of the scaling constants.
656 (rt->rt_rmx.rmx_rtt + i) / 2;
658 rt->rt_rmx.rmx_rtt = i;
659 tcpstat.tcps_cachedrtt++;
661 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
663 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
664 if (rt->rt_rmx.rmx_rttvar && i)
665 rt->rt_rmx.rmx_rttvar =
666 (rt->rt_rmx.rmx_rttvar + i) / 2;
668 rt->rt_rmx.rmx_rttvar = i;
669 tcpstat.tcps_cachedrttvar++;
672 * The old comment here said:
673 * update the pipelimit (ssthresh) if it has been updated
674 * already or if a pipesize was specified & the threshhold
675 * got below half the pipesize. I.e., wait for bad news
676 * before we start updating, then update on both good
679 * But we want to save the ssthresh even if no pipesize is
680 * specified explicitly in the route, because such
681 * connections still have an implicit pipesize specified
682 * by the global tcp_sendspace. In the absence of a reliable
683 * way to calculate the pipesize, it will have to do.
685 i = tp->snd_ssthresh;
686 if (rt->rt_rmx.rmx_sendpipe != 0)
687 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
689 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
690 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
691 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
694 * convert the limit from user data bytes to
695 * packets then to packet data bytes.
697 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
700 i *= (u_long)(tp->t_maxseg +
702 (isipv6 ? sizeof (struct ip6_hdr) +
703 sizeof (struct tcphdr) :
705 sizeof (struct tcpiphdr)
710 if (rt->rt_rmx.rmx_ssthresh)
711 rt->rt_rmx.rmx_ssthresh =
712 (rt->rt_rmx.rmx_ssthresh + i) / 2;
714 rt->rt_rmx.rmx_ssthresh = i;
715 tcpstat.tcps_cachedssthresh++;
719 /* free the reassembly queue, if any */
720 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
721 LIST_REMOVE(q, tqe_q);
725 inp->inp_ppcb = NULL;
726 soisdisconnected(so);
728 if (INP_CHECK_SOCKAF(so, AF_INET6))
733 tcpstat.tcps_closed++;
734 return ((struct tcpcb *)0);
744 struct tseg_qent *te;
747 * Walk the tcpbs, if existing, and flush the reassembly queue,
749 * XXX: The "Net/3" implementation doesn't imply that the TCP
750 * reassembly queue should be flushed, but in a situation
751 * where we're really low on mbufs, this is potentially
754 INP_INFO_RLOCK(&tcbinfo);
755 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
757 if ((tcpb = intotcpcb(inpb))) {
758 while ((te = LIST_FIRST(&tcpb->t_segq))
760 LIST_REMOVE(te, tqe_q);
767 INP_INFO_RUNLOCK(&tcbinfo);
772 * Notify a tcp user of an asynchronous error;
773 * store error as soft error, but wake up user
774 * (for now, won't do anything until can select for soft error).
776 * Do not wake up user since there currently is no mechanism for
777 * reporting soft errors (yet - a kqueue filter may be added).
779 static struct inpcb *
780 tcp_notify(inp, error)
784 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
787 * Ignore some errors if we are hooked up.
788 * If connection hasn't completed, has retransmitted several times,
789 * and receives a second error, give up now. This is better
790 * than waiting a long time to establish a connection that
791 * can never complete.
793 if (tp->t_state == TCPS_ESTABLISHED &&
794 (error == EHOSTUNREACH || error == ENETUNREACH ||
795 error == EHOSTDOWN)) {
797 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
800 return (struct inpcb *)0;
802 tp->t_softerror = error;
806 wakeup((caddr_t) &so->so_timeo);
813 tcp_pcblist(SYSCTL_HANDLER_ARGS)
816 struct inpcb *inp, **inp_list;
821 * The process of preparing the TCB list is too time-consuming and
822 * resource-intensive to repeat twice on every request.
824 if (req->oldptr == 0) {
825 n = tcbinfo.ipi_count;
826 req->oldidx = 2 * (sizeof xig)
827 + (n + n/8) * sizeof(struct xtcpcb);
831 if (req->newptr != 0)
835 * OK, now we're committed to doing something.
838 INP_INFO_RLOCK(&tcbinfo);
839 gencnt = tcbinfo.ipi_gencnt;
840 n = tcbinfo.ipi_count;
841 INP_INFO_RUNLOCK(&tcbinfo);
844 sysctl_wire_old_buffer(req, 2 * (sizeof xig)
845 + n * sizeof(struct xtcpcb));
847 xig.xig_len = sizeof xig;
849 xig.xig_gen = gencnt;
850 xig.xig_sogen = so_gencnt;
851 error = SYSCTL_OUT(req, &xig, sizeof xig);
855 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
860 INP_INFO_RLOCK(&tcbinfo);
861 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
862 inp = LIST_NEXT(inp, inp_list)) {
864 if (inp->inp_gencnt <= gencnt &&
865 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
869 INP_INFO_RUNLOCK(&tcbinfo);
874 for (i = 0; i < n; i++) {
877 if (inp->inp_gencnt <= gencnt) {
880 xt.xt_len = sizeof xt;
881 /* XXX should avoid extra copy */
882 bcopy(inp, &xt.xt_inp, sizeof *inp);
883 inp_ppcb = inp->inp_ppcb;
884 if (inp_ppcb != NULL)
885 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
887 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
889 sotoxsocket(inp->inp_socket, &xt.xt_socket);
890 error = SYSCTL_OUT(req, &xt, sizeof xt);
896 * Give the user an updated idea of our state.
897 * If the generation differs from what we told
898 * her before, she knows that something happened
899 * while we were processing this request, and it
900 * might be necessary to retry.
903 INP_INFO_RLOCK(&tcbinfo);
904 xig.xig_gen = tcbinfo.ipi_gencnt;
905 xig.xig_sogen = so_gencnt;
906 xig.xig_count = tcbinfo.ipi_count;
907 INP_INFO_RUNLOCK(&tcbinfo);
909 error = SYSCTL_OUT(req, &xig, sizeof xig);
911 free(inp_list, M_TEMP);
915 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
916 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
919 tcp_getcred(SYSCTL_HANDLER_ARGS)
922 struct sockaddr_in addrs[2];
926 error = suser_cred(req->td->td_ucred, PRISON_ROOT);
929 error = SYSCTL_IN(req, addrs, sizeof(addrs));
933 INP_INFO_RLOCK(&tcbinfo);
934 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
935 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
941 if (inp->inp_socket == NULL) {
945 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
948 cru2x(inp->inp_socket->so_cred, &xuc);
952 INP_INFO_RUNLOCK(&tcbinfo);
955 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
959 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
960 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
961 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
965 tcp6_getcred(SYSCTL_HANDLER_ARGS)
968 struct sockaddr_in6 addrs[2];
970 int error, s, mapped = 0;
972 error = suser_cred(req->td->td_ucred, PRISON_ROOT);
975 error = SYSCTL_IN(req, addrs, sizeof(addrs));
978 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
979 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
985 INP_INFO_RLOCK(&tcbinfo);
987 inp = in_pcblookup_hash(&tcbinfo,
988 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
990 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
994 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
996 &addrs[0].sin6_addr, addrs[0].sin6_port,
1003 if (inp->inp_socket == NULL) {
1007 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1010 cru2x(inp->inp_socket->so_cred, &xuc);
1014 INP_INFO_RUNLOCK(&tcbinfo);
1017 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1021 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1022 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1023 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1028 tcp_ctlinput(cmd, sa, vip)
1030 struct sockaddr *sa;
1033 struct ip *ip = vip;
1035 struct in_addr faddr;
1038 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1042 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1043 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1046 if (cmd == PRC_QUENCH)
1047 notify = tcp_quench;
1048 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1049 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1050 notify = tcp_drop_syn_sent;
1051 else if (cmd == PRC_MSGSIZE)
1052 notify = tcp_mtudisc;
1053 else if (PRC_IS_REDIRECT(cmd)) {
1055 notify = in_rtchange;
1056 } else if (cmd == PRC_HOSTDEAD)
1058 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
1062 th = (struct tcphdr *)((caddr_t)ip
1063 + (IP_VHL_HL(ip->ip_vhl) << 2));
1064 INP_INFO_WLOCK(&tcbinfo);
1065 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1066 ip->ip_src, th->th_sport, 0, NULL);
1069 if (inp->inp_socket != NULL) {
1070 icmp_seq = htonl(th->th_seq);
1071 tp = intotcpcb(inp);
1072 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1073 SEQ_LT(icmp_seq, tp->snd_max))
1074 inp = (*notify)(inp, inetctlerrmap[cmd]);
1079 struct in_conninfo inc;
1081 inc.inc_fport = th->th_dport;
1082 inc.inc_lport = th->th_sport;
1083 inc.inc_faddr = faddr;
1084 inc.inc_laddr = ip->ip_src;
1088 syncache_unreach(&inc, th);
1090 INP_INFO_WUNLOCK(&tcbinfo);
1093 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1098 tcp6_ctlinput(cmd, sa, d)
1100 struct sockaddr *sa;
1104 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1105 struct ip6_hdr *ip6;
1107 struct ip6ctlparam *ip6cp = NULL;
1108 const struct sockaddr_in6 *sa6_src = NULL;
1110 struct tcp_portonly {
1115 if (sa->sa_family != AF_INET6 ||
1116 sa->sa_len != sizeof(struct sockaddr_in6))
1119 if (cmd == PRC_QUENCH)
1120 notify = tcp_quench;
1121 else if (cmd == PRC_MSGSIZE)
1122 notify = tcp_mtudisc;
1123 else if (!PRC_IS_REDIRECT(cmd) &&
1124 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1127 /* if the parameter is from icmp6, decode it. */
1129 ip6cp = (struct ip6ctlparam *)d;
1131 ip6 = ip6cp->ip6c_ip6;
1132 off = ip6cp->ip6c_off;
1133 sa6_src = ip6cp->ip6c_src;
1137 off = 0; /* fool gcc */
1142 struct in_conninfo inc;
1144 * XXX: We assume that when IPV6 is non NULL,
1145 * M and OFF are valid.
1148 /* check if we can safely examine src and dst ports */
1149 if (m->m_pkthdr.len < off + sizeof(*thp))
1152 bzero(&th, sizeof(th));
1153 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1155 in6_pcbnotify(&tcb, sa, th.th_dport,
1156 (struct sockaddr *)ip6cp->ip6c_src,
1157 th.th_sport, cmd, notify);
1159 inc.inc_fport = th.th_dport;
1160 inc.inc_lport = th.th_sport;
1161 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1162 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1164 syncache_unreach(&inc, &th);
1166 in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src,
1173 * Following is where TCP initial sequence number generation occurs.
1175 * There are two places where we must use initial sequence numbers:
1176 * 1. In SYN-ACK packets.
1177 * 2. In SYN packets.
1179 * All ISNs for SYN-ACK packets are generated by the syncache. See
1180 * tcp_syncache.c for details.
1182 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1183 * depends on this property. In addition, these ISNs should be
1184 * unguessable so as to prevent connection hijacking. To satisfy
1185 * the requirements of this situation, the algorithm outlined in
1186 * RFC 1948 is used to generate sequence numbers.
1188 * Implementation details:
1190 * Time is based off the system timer, and is corrected so that it
1191 * increases by one megabyte per second. This allows for proper
1192 * recycling on high speed LANs while still leaving over an hour
1195 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1196 * between seeding of isn_secret. This is normally set to zero,
1197 * as reseeding should not be necessary.
1201 #define ISN_BYTES_PER_SECOND 1048576
1203 u_char isn_secret[32];
1204 int isn_last_reseed;
1211 u_int32_t md5_buffer[4];
1214 /* Seed if this is the first use, reseed if requested. */
1215 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1216 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1218 read_random(&isn_secret, sizeof(isn_secret));
1219 isn_last_reseed = ticks;
1222 /* Compute the md5 hash and return the ISN. */
1224 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1225 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1227 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1228 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1229 sizeof(struct in6_addr));
1230 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1231 sizeof(struct in6_addr));
1235 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1236 sizeof(struct in_addr));
1237 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1238 sizeof(struct in_addr));
1240 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1241 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1242 new_isn = (tcp_seq) md5_buffer[0];
1243 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
1248 * When a source quench is received, close congestion window
1249 * to one segment. We will gradually open it again as we proceed.
1252 tcp_quench(inp, errno)
1256 struct tcpcb *tp = intotcpcb(inp);
1259 tp->snd_cwnd = tp->t_maxseg;
1264 * When a specific ICMP unreachable message is received and the
1265 * connection state is SYN-SENT, drop the connection. This behavior
1266 * is controlled by the icmp_may_rst sysctl.
1269 tcp_drop_syn_sent(inp, errno)
1273 struct tcpcb *tp = intotcpcb(inp);
1275 if (tp && tp->t_state == TCPS_SYN_SENT) {
1276 tcp_drop(tp, errno);
1277 return (struct inpcb *)0;
1283 * When `need fragmentation' ICMP is received, update our idea of the MSS
1284 * based on the new value in the route. Also nudge TCP to send something,
1285 * since we know the packet we just sent was dropped.
1286 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1289 tcp_mtudisc(inp, errno)
1293 struct tcpcb *tp = intotcpcb(inp);
1295 struct rmxp_tao *taop;
1296 struct socket *so = inp->inp_socket;
1300 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1306 rt = tcp_rtlookup6(&inp->inp_inc);
1309 rt = tcp_rtlookup(&inp->inp_inc);
1310 if (!rt || !rt->rt_rmx.rmx_mtu) {
1311 tp->t_maxopd = tp->t_maxseg =
1313 isipv6 ? tcp_v6mssdflt :
1318 taop = rmx_taop(rt->rt_rmx);
1319 offered = taop->tao_mssopt;
1320 mss = rt->rt_rmx.rmx_mtu -
1323 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1325 sizeof(struct tcpiphdr)
1332 mss = min(mss, offered);
1334 * XXX - The above conditional probably violates the TCP
1335 * spec. The problem is that, since we don't know the
1336 * other end's MSS, we are supposed to use a conservative
1337 * default. But, if we do that, then MTU discovery will
1338 * never actually take place, because the conservative
1339 * default is much less than the MTUs typically seen
1340 * on the Internet today. For the moment, we'll sweep
1341 * this under the carpet.
1343 * The conservative default might not actually be a problem
1344 * if the only case this occurs is when sending an initial
1345 * SYN with options and data to a host we've never talked
1346 * to before. Then, they will reply with an MSS value which
1347 * will get recorded and the new parameters should get
1348 * recomputed. For Further Study.
1350 if (tp->t_maxopd <= mss)
1354 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1355 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1356 mss -= TCPOLEN_TSTAMP_APPA;
1357 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1358 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1359 mss -= TCPOLEN_CC_APPA;
1360 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1362 mss &= ~(MCLBYTES-1);
1365 mss = mss / MCLBYTES * MCLBYTES;
1367 if (so->so_snd.sb_hiwat < mss)
1368 mss = so->so_snd.sb_hiwat;
1372 tcpstat.tcps_mturesent++;
1374 tp->snd_nxt = tp->snd_una;
1381 * Look-up the routing entry to the peer of this inpcb. If no route
1382 * is found and it cannot be allocated the return NULL. This routine
1383 * is called by TCP routines that access the rmx structure and by tcp_mss
1384 * to get the interface MTU.
1388 struct in_conninfo *inc;
1393 ro = &inc->inc_route;
1395 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1396 /* No route yet, so try to acquire one */
1397 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1398 ro->ro_dst.sa_family = AF_INET;
1399 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1400 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1412 struct in_conninfo *inc;
1414 struct route_in6 *ro6;
1417 ro6 = &inc->inc6_route;
1419 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1420 /* No route yet, so try to acquire one */
1421 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1422 ro6->ro_dst.sin6_family = AF_INET6;
1423 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1424 ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1425 rtalloc((struct route *)ro6);
1434 /* compute ESP/AH header size for TCP, including outer IP header. */
1436 ipsec_hdrsiz_tcp(tp)
1444 struct ip6_hdr *ip6;
1448 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1450 MGETHDR(m, M_DONTWAIT, MT_DATA);
1455 if ((inp->inp_vflag & INP_IPV6) != 0) {
1456 ip6 = mtod(m, struct ip6_hdr *);
1457 th = (struct tcphdr *)(ip6 + 1);
1458 m->m_pkthdr.len = m->m_len =
1459 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1460 tcp_fillheaders(tp, ip6, th);
1461 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1465 ip = mtod(m, struct ip *);
1466 th = (struct tcphdr *)(ip + 1);
1467 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1468 tcp_fillheaders(tp, ip, th);
1469 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1478 * Return a pointer to the cached information about the remote host.
1479 * The cached information is stored in the protocol specific part of
1480 * the route metrics.
1483 tcp_gettaocache(inc)
1484 struct in_conninfo *inc;
1489 if (inc->inc_isipv6)
1490 rt = tcp_rtlookup6(inc);
1493 rt = tcp_rtlookup(inc);
1495 /* Make sure this is a host route and is up. */
1497 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1500 return rmx_taop(rt->rt_rmx);
1504 * Clear all the TAO cache entries, called from tcp_init.
1507 * This routine is just an empty one, because we assume that the routing
1508 * routing tables are initialized at the same time when TCP, so there is
1509 * nothing in the cache left over.