Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.57
      1  1.57   thorpej /*	$NetBSD: tcp_subr.c,v 1.57 1998/08/02 00:36:19 thorpej Exp $	*/
      2  1.41   thorpej 
      3  1.41   thorpej /*-
      4  1.41   thorpej  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
      5  1.41   thorpej  * All rights reserved.
      6  1.41   thorpej  *
      7  1.41   thorpej  * This code is derived from software contributed to The NetBSD Foundation
      8  1.41   thorpej  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
      9  1.41   thorpej  * Facility, NASA Ames Research Center.
     10  1.41   thorpej  *
     11  1.41   thorpej  * Redistribution and use in source and binary forms, with or without
     12  1.41   thorpej  * modification, are permitted provided that the following conditions
     13  1.41   thorpej  * are met:
     14  1.41   thorpej  * 1. Redistributions of source code must retain the above copyright
     15  1.41   thorpej  *    notice, this list of conditions and the following disclaimer.
     16  1.41   thorpej  * 2. Redistributions in binary form must reproduce the above copyright
     17  1.41   thorpej  *    notice, this list of conditions and the following disclaimer in the
     18  1.41   thorpej  *    documentation and/or other materials provided with the distribution.
     19  1.41   thorpej  * 3. All advertising materials mentioning features or use of this software
     20  1.41   thorpej  *    must display the following acknowledgement:
     21  1.41   thorpej  *	This product includes software developed by the NetBSD
     22  1.41   thorpej  *	Foundation, Inc. and its contributors.
     23  1.41   thorpej  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  1.41   thorpej  *    contributors may be used to endorse or promote products derived
     25  1.41   thorpej  *    from this software without specific prior written permission.
     26  1.41   thorpej  *
     27  1.41   thorpej  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  1.41   thorpej  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  1.41   thorpej  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  1.41   thorpej  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  1.41   thorpej  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  1.41   thorpej  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  1.41   thorpej  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  1.41   thorpej  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  1.41   thorpej  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  1.41   thorpej  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  1.41   thorpej  * POSSIBILITY OF SUCH DAMAGE.
     38  1.41   thorpej  */
     39  1.11       cgd 
     40   1.1       cgd /*
     41  1.38   thorpej  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
     42  1.10   mycroft  *	The Regents of the University of California.  All rights reserved.
     43   1.1       cgd  *
     44   1.1       cgd  * Redistribution and use in source and binary forms, with or without
     45   1.1       cgd  * modification, are permitted provided that the following conditions
     46   1.1       cgd  * are met:
     47   1.1       cgd  * 1. Redistributions of source code must retain the above copyright
     48   1.1       cgd  *    notice, this list of conditions and the following disclaimer.
     49   1.1       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     50   1.1       cgd  *    notice, this list of conditions and the following disclaimer in the
     51   1.1       cgd  *    documentation and/or other materials provided with the distribution.
     52   1.1       cgd  * 3. All advertising materials mentioning features or use of this software
     53   1.1       cgd  *    must display the following acknowledgement:
     54   1.1       cgd  *	This product includes software developed by the University of
     55   1.1       cgd  *	California, Berkeley and its contributors.
     56   1.1       cgd  * 4. Neither the name of the University nor the names of its contributors
     57   1.1       cgd  *    may be used to endorse or promote products derived from this software
     58   1.1       cgd  *    without specific prior written permission.
     59   1.1       cgd  *
     60   1.1       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     61   1.1       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     62   1.1       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     63   1.1       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     64   1.1       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     65   1.1       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     66   1.1       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     67   1.1       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     68   1.1       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     69   1.1       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     70   1.1       cgd  * SUCH DAMAGE.
     71   1.1       cgd  *
     72  1.38   thorpej  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
     73   1.1       cgd  */
     74   1.1       cgd 
     75  1.39    scottr #include "opt_tcp_compat_42.h"
     76  1.30  explorer #include "rnd.h"
     77  1.30  explorer 
     78   1.5   mycroft #include <sys/param.h>
     79  1.10   mycroft #include <sys/proc.h>
     80   1.5   mycroft #include <sys/systm.h>
     81   1.5   mycroft #include <sys/malloc.h>
     82   1.5   mycroft #include <sys/mbuf.h>
     83   1.5   mycroft #include <sys/socket.h>
     84   1.5   mycroft #include <sys/socketvar.h>
     85   1.5   mycroft #include <sys/protosw.h>
     86   1.5   mycroft #include <sys/errno.h>
     87  1.27   thorpej #include <sys/kernel.h>
     88  1.57   thorpej #include <sys/pool.h>
     89  1.30  explorer #if NRND > 0
     90  1.29  explorer #include <sys/rnd.h>
     91  1.30  explorer #endif
     92   1.1       cgd 
     93   1.5   mycroft #include <net/route.h>
     94   1.5   mycroft #include <net/if.h>
     95   1.1       cgd 
     96   1.5   mycroft #include <netinet/in.h>
     97   1.5   mycroft #include <netinet/in_systm.h>
     98   1.5   mycroft #include <netinet/ip.h>
     99   1.5   mycroft #include <netinet/in_pcb.h>
    100   1.5   mycroft #include <netinet/ip_var.h>
    101   1.5   mycroft #include <netinet/ip_icmp.h>
    102   1.5   mycroft #include <netinet/tcp.h>
    103   1.5   mycroft #include <netinet/tcp_fsm.h>
    104   1.5   mycroft #include <netinet/tcp_seq.h>
    105   1.5   mycroft #include <netinet/tcp_timer.h>
    106   1.5   mycroft #include <netinet/tcp_var.h>
    107   1.5   mycroft #include <netinet/tcpip.h>
    108   1.1       cgd 
    109   1.1       cgd /* patchable/settable parameters for tcp */
    110   1.1       cgd int 	tcp_mssdflt = TCP_MSS;
    111   1.1       cgd int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
    112  1.49      matt int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
    113  1.49      matt int	tcp_do_sack = 1;	/* selective acknowledgement */
    114  1.49      matt int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
    115  1.49      matt int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
    116  1.50   thorpej int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
    117  1.36   thorpej int	tcp_init_win = 1;
    118  1.47       kml int	tcp_mss_ifmtu = 0;
    119  1.48   thorpej #ifdef TCP_COMPAT_42
    120  1.48   thorpej int	tcp_compat_42 = 1;
    121  1.48   thorpej #else
    122  1.48   thorpej int	tcp_compat_42 = 0;
    123  1.48   thorpej #endif
    124   1.1       cgd 
    125  1.21   mycroft #ifndef TCBHASHSIZE
    126  1.21   mycroft #define	TCBHASHSIZE	128
    127  1.21   mycroft #endif
    128  1.21   mycroft int	tcbhashsize = TCBHASHSIZE;
    129   1.1       cgd 
    130  1.35   thorpej int	tcp_freeq __P((struct tcpcb *));
    131  1.35   thorpej 
    132  1.57   thorpej struct pool tcpcb_pool;
    133  1.57   thorpej 
    134   1.1       cgd /*
    135   1.1       cgd  * Tcp initialization
    136   1.1       cgd  */
    137   1.7   mycroft void
    138   1.1       cgd tcp_init()
    139   1.1       cgd {
    140   1.1       cgd 
    141  1.57   thorpej 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
    142  1.57   thorpej 	    0, NULL, NULL, M_PCB);
    143  1.24   mycroft 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
    144  1.37   thorpej 	LIST_INIT(&tcp_delacks);
    145   1.1       cgd 	if (max_protohdr < sizeof(struct tcpiphdr))
    146   1.1       cgd 		max_protohdr = sizeof(struct tcpiphdr);
    147   1.1       cgd 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
    148   1.1       cgd 		panic("tcp_init");
    149  1.52   thorpej 
    150  1.52   thorpej 	/* Initialize the compressed state engine. */
    151  1.52   thorpej 	syn_cache_init();
    152   1.1       cgd }
    153   1.1       cgd 
    154   1.1       cgd /*
    155   1.1       cgd  * Create template to be used to send tcp packets on a connection.
    156   1.1       cgd  * Call after host entry created, allocates an mbuf and fills
    157   1.1       cgd  * in a skeletal tcp/ip header, minimizing the amount of work
    158   1.1       cgd  * necessary when the connection is used.
    159   1.1       cgd  */
    160   1.1       cgd struct tcpiphdr *
    161   1.1       cgd tcp_template(tp)
    162   1.1       cgd 	struct tcpcb *tp;
    163   1.1       cgd {
    164   1.1       cgd 	register struct inpcb *inp = tp->t_inpcb;
    165   1.1       cgd 	register struct tcpiphdr *n;
    166   1.1       cgd 
    167   1.1       cgd 	if ((n = tp->t_template) == 0) {
    168  1.26   thorpej 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    169  1.26   thorpej 		    M_MBUF, M_NOWAIT);
    170  1.26   thorpej 		if (n == NULL)
    171   1.1       cgd 			return (0);
    172   1.1       cgd 	}
    173  1.20       cgd 	bzero(n->ti_x1, sizeof n->ti_x1);
    174   1.1       cgd 	n->ti_pr = IPPROTO_TCP;
    175   1.1       cgd 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    176   1.1       cgd 	n->ti_src = inp->inp_laddr;
    177   1.1       cgd 	n->ti_dst = inp->inp_faddr;
    178   1.1       cgd 	n->ti_sport = inp->inp_lport;
    179   1.1       cgd 	n->ti_dport = inp->inp_fport;
    180   1.1       cgd 	n->ti_seq = 0;
    181   1.1       cgd 	n->ti_ack = 0;
    182   1.1       cgd 	n->ti_x2 = 0;
    183   1.1       cgd 	n->ti_off = 5;
    184   1.1       cgd 	n->ti_flags = 0;
    185   1.1       cgd 	n->ti_win = 0;
    186   1.1       cgd 	n->ti_sum = 0;
    187   1.1       cgd 	n->ti_urp = 0;
    188   1.1       cgd 	return (n);
    189   1.1       cgd }
    190   1.1       cgd 
    191   1.1       cgd /*
    192   1.1       cgd  * Send a single message to the TCP at address specified by
    193   1.1       cgd  * the given TCP/IP header.  If m == 0, then we make a copy
    194   1.1       cgd  * of the tcpiphdr at ti and send directly to the addressed host.
    195   1.1       cgd  * This is used to force keep alive messages out using the TCP
    196   1.1       cgd  * template for a connection tp->t_template.  If flags are given
    197   1.1       cgd  * then we send a message back to the TCP which originated the
    198   1.1       cgd  * segment ti, and discard the mbuf containing it and any other
    199   1.1       cgd  * attached mbufs.
    200   1.1       cgd  *
    201   1.1       cgd  * In any case the ack and sequence number of the transmitted
    202   1.1       cgd  * segment are as specified by the parameters.
    203   1.1       cgd  */
    204  1.27   thorpej int
    205   1.1       cgd tcp_respond(tp, ti, m, ack, seq, flags)
    206   1.1       cgd 	struct tcpcb *tp;
    207   1.1       cgd 	register struct tcpiphdr *ti;
    208   1.1       cgd 	register struct mbuf *m;
    209   1.1       cgd 	tcp_seq ack, seq;
    210   1.1       cgd 	int flags;
    211   1.1       cgd {
    212   1.1       cgd 	register int tlen;
    213   1.1       cgd 	int win = 0;
    214   1.1       cgd 	struct route *ro = 0;
    215   1.1       cgd 
    216   1.1       cgd 	if (tp) {
    217   1.1       cgd 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    218   1.1       cgd 		ro = &tp->t_inpcb->inp_route;
    219   1.1       cgd 	}
    220   1.1       cgd 	if (m == 0) {
    221   1.1       cgd 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    222   1.1       cgd 		if (m == NULL)
    223  1.27   thorpej 			return (ENOBUFS);
    224  1.48   thorpej 
    225  1.48   thorpej 		if (tcp_compat_42)
    226  1.48   thorpej 			tlen = 1;
    227  1.48   thorpej 		else
    228  1.48   thorpej 			tlen = 0;
    229  1.48   thorpej 
    230   1.1       cgd 		m->m_data += max_linkhdr;
    231   1.1       cgd 		*mtod(m, struct tcpiphdr *) = *ti;
    232   1.1       cgd 		ti = mtod(m, struct tcpiphdr *);
    233   1.1       cgd 		flags = TH_ACK;
    234   1.1       cgd 	} else {
    235   1.1       cgd 		m_freem(m->m_next);
    236   1.1       cgd 		m->m_next = 0;
    237   1.1       cgd 		m->m_data = (caddr_t)ti;
    238   1.1       cgd 		m->m_len = sizeof (struct tcpiphdr);
    239   1.1       cgd 		tlen = 0;
    240  1.10   mycroft #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    241  1.13       cgd 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    242  1.13       cgd 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    243   1.1       cgd #undef xchg
    244   1.1       cgd 	}
    245  1.20       cgd 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    246   1.1       cgd 	ti->ti_seq = htonl(seq);
    247   1.1       cgd 	ti->ti_ack = htonl(ack);
    248   1.1       cgd 	ti->ti_x2 = 0;
    249  1.27   thorpej 	if ((flags & TH_SYN) == 0) {
    250  1.27   thorpej 		if (tp)
    251  1.27   thorpej 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    252  1.27   thorpej 		else
    253  1.27   thorpej 			ti->ti_win = htons((u_int16_t)win);
    254  1.27   thorpej 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    255  1.27   thorpej 		tlen += sizeof (struct tcphdr);
    256  1.27   thorpej 	} else
    257  1.27   thorpej 		tlen += ti->ti_off << 2;
    258  1.27   thorpej 	ti->ti_len = htons((u_int16_t)tlen);
    259  1.27   thorpej 	tlen += sizeof (struct ip);
    260  1.27   thorpej 	m->m_len = tlen;
    261  1.27   thorpej 	m->m_pkthdr.len = tlen;
    262  1.27   thorpej 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    263   1.1       cgd 	ti->ti_flags = flags;
    264   1.1       cgd 	ti->ti_urp = 0;
    265  1.10   mycroft 	ti->ti_sum = 0;
    266   1.1       cgd 	ti->ti_sum = in_cksum(m, tlen);
    267   1.1       cgd 	((struct ip *)ti)->ip_len = tlen;
    268  1.10   mycroft 	((struct ip *)ti)->ip_ttl = ip_defttl;
    269  1.27   thorpej 	return ip_output(m, NULL, ro, 0, NULL);
    270   1.1       cgd }
    271   1.1       cgd 
    272   1.1       cgd /*
    273   1.1       cgd  * Create a new TCP control block, making an
    274   1.1       cgd  * empty reassembly queue and hooking it to the argument
    275   1.1       cgd  * protocol control block.
    276   1.1       cgd  */
    277   1.1       cgd struct tcpcb *
    278   1.1       cgd tcp_newtcpcb(inp)
    279   1.1       cgd 	struct inpcb *inp;
    280   1.1       cgd {
    281   1.1       cgd 	register struct tcpcb *tp;
    282   1.1       cgd 
    283  1.57   thorpej 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
    284  1.10   mycroft 	if (tp == NULL)
    285  1.57   thorpej 		return (NULL);
    286  1.23   mycroft 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    287  1.20       cgd 	LIST_INIT(&tp->segq);
    288  1.49      matt 	LIST_INIT(&tp->timeq);
    289  1.33       kml 	tp->t_peermss = tcp_mssdflt;
    290  1.28   thorpej 	tp->t_ourmss = tcp_mssdflt;
    291  1.33       kml 	tp->t_segsz = tcp_mssdflt;
    292   1.1       cgd 
    293  1.49      matt 	tp->t_flags = 0;
    294  1.49      matt 	if (tcp_do_rfc1323 && tcp_do_win_scale)
    295  1.49      matt 		tp->t_flags |= TF_REQ_SCALE;
    296  1.49      matt 	if (tcp_do_rfc1323 && tcp_do_timestamps)
    297  1.49      matt 		tp->t_flags |= TF_REQ_TSTMP;
    298  1.49      matt 	if (tcp_do_sack == 2)
    299  1.49      matt 		tp->t_flags |= TF_WILL_SACK;
    300  1.49      matt 	else if (tcp_do_sack == 1)
    301  1.49      matt 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
    302  1.49      matt 	tp->t_flags |= TF_CANT_TXSACK;
    303   1.1       cgd 	tp->t_inpcb = inp;
    304   1.1       cgd 	/*
    305   1.1       cgd 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    306   1.1       cgd 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    307   1.1       cgd 	 * reasonable initial retransmit time.
    308   1.1       cgd 	 */
    309   1.1       cgd 	tp->t_srtt = TCPTV_SRTTBASE;
    310  1.15   mycroft 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    311   1.1       cgd 	tp->t_rttmin = TCPTV_MIN;
    312  1.15   mycroft 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    313   1.1       cgd 	    TCPTV_MIN, TCPTV_REXMTMAX);
    314  1.10   mycroft 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    315  1.10   mycroft 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    316  1.10   mycroft 	inp->inp_ip.ip_ttl = ip_defttl;
    317   1.1       cgd 	inp->inp_ppcb = (caddr_t)tp;
    318   1.1       cgd 	return (tp);
    319   1.1       cgd }
    320   1.1       cgd 
    321   1.1       cgd /*
    322   1.1       cgd  * Drop a TCP connection, reporting
    323   1.1       cgd  * the specified error.  If connection is synchronized,
    324   1.1       cgd  * then send a RST to peer.
    325   1.1       cgd  */
    326   1.1       cgd struct tcpcb *
    327   1.1       cgd tcp_drop(tp, errno)
    328   1.1       cgd 	register struct tcpcb *tp;
    329   1.1       cgd 	int errno;
    330   1.1       cgd {
    331   1.1       cgd 	struct socket *so = tp->t_inpcb->inp_socket;
    332   1.1       cgd 
    333   1.1       cgd 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    334   1.1       cgd 		tp->t_state = TCPS_CLOSED;
    335   1.1       cgd 		(void) tcp_output(tp);
    336   1.1       cgd 		tcpstat.tcps_drops++;
    337   1.1       cgd 	} else
    338   1.1       cgd 		tcpstat.tcps_conndrops++;
    339   1.1       cgd 	if (errno == ETIMEDOUT && tp->t_softerror)
    340   1.1       cgd 		errno = tp->t_softerror;
    341   1.1       cgd 	so->so_error = errno;
    342   1.1       cgd 	return (tcp_close(tp));
    343   1.1       cgd }
    344   1.1       cgd 
    345   1.1       cgd /*
    346   1.1       cgd  * Close a TCP control block:
    347   1.1       cgd  *	discard all space held by the tcp
    348   1.1       cgd  *	discard internet protocol block
    349   1.1       cgd  *	wake up any sleepers
    350   1.1       cgd  */
    351   1.1       cgd struct tcpcb *
    352   1.1       cgd tcp_close(tp)
    353   1.1       cgd 	register struct tcpcb *tp;
    354   1.1       cgd {
    355   1.1       cgd 	struct inpcb *inp = tp->t_inpcb;
    356   1.1       cgd 	struct socket *so = inp->inp_socket;
    357   1.1       cgd #ifdef RTV_RTT
    358   1.1       cgd 	register struct rtentry *rt;
    359   1.1       cgd 
    360   1.1       cgd 	/*
    361   1.1       cgd 	 * If we sent enough data to get some meaningful characteristics,
    362  1.10   mycroft 	 * save them in the routing entry.  'Enough' is arbitrarily
    363   1.1       cgd 	 * defined as the sendpipesize (default 4K) * 16.  This would
    364   1.1       cgd 	 * give us 16 rtt samples assuming we only get one sample per
    365   1.1       cgd 	 * window (the usual case on a long haul net).  16 samples is
    366   1.1       cgd 	 * enough for the srtt filter to converge to within 5% of the correct
    367   1.1       cgd 	 * value; fewer samples and we could save a very bogus rtt.
    368   1.1       cgd 	 *
    369   1.1       cgd 	 * Don't update the default route's characteristics and don't
    370   1.1       cgd 	 * update anything that the user "locked".
    371   1.1       cgd 	 */
    372   1.1       cgd 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    373   1.1       cgd 	    (rt = inp->inp_route.ro_rt) &&
    374  1.23   mycroft 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    375  1.22  christos 		register u_long i = 0;
    376   1.1       cgd 
    377   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    378   1.1       cgd 			i = tp->t_srtt *
    379  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    380   1.1       cgd 			if (rt->rt_rmx.rmx_rtt && i)
    381   1.1       cgd 				/*
    382   1.1       cgd 				 * filter this update to half the old & half
    383   1.1       cgd 				 * the new values, converting scale.
    384   1.1       cgd 				 * See route.h and tcp_var.h for a
    385   1.1       cgd 				 * description of the scaling constants.
    386   1.1       cgd 				 */
    387   1.1       cgd 				rt->rt_rmx.rmx_rtt =
    388   1.1       cgd 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    389   1.1       cgd 			else
    390   1.1       cgd 				rt->rt_rmx.rmx_rtt = i;
    391   1.1       cgd 		}
    392   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    393   1.1       cgd 			i = tp->t_rttvar *
    394  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    395   1.1       cgd 			if (rt->rt_rmx.rmx_rttvar && i)
    396   1.1       cgd 				rt->rt_rmx.rmx_rttvar =
    397   1.1       cgd 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    398   1.1       cgd 			else
    399   1.1       cgd 				rt->rt_rmx.rmx_rttvar = i;
    400   1.1       cgd 		}
    401   1.1       cgd 		/*
    402   1.1       cgd 		 * update the pipelimit (ssthresh) if it has been updated
    403   1.1       cgd 		 * already or if a pipesize was specified & the threshhold
    404   1.1       cgd 		 * got below half the pipesize.  I.e., wait for bad news
    405   1.1       cgd 		 * before we start updating, then update on both good
    406   1.1       cgd 		 * and bad news.
    407   1.1       cgd 		 */
    408  1.22  christos 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    409  1.22  christos 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    410   1.1       cgd 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    411   1.1       cgd 			/*
    412   1.1       cgd 			 * convert the limit from user data bytes to
    413   1.1       cgd 			 * packets then to packet data bytes.
    414   1.1       cgd 			 */
    415  1.33       kml 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
    416   1.1       cgd 			if (i < 2)
    417   1.1       cgd 				i = 2;
    418  1.33       kml 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
    419   1.1       cgd 			if (rt->rt_rmx.rmx_ssthresh)
    420   1.1       cgd 				rt->rt_rmx.rmx_ssthresh =
    421   1.1       cgd 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    422   1.1       cgd 			else
    423   1.1       cgd 				rt->rt_rmx.rmx_ssthresh = i;
    424   1.1       cgd 		}
    425   1.1       cgd 	}
    426   1.9   mycroft #endif /* RTV_RTT */
    427   1.1       cgd 	/* free the reassembly queue, if any */
    428  1.35   thorpej 	(void) tcp_freeq(tp);
    429  1.40    mellon 	TCP_CLEAR_DELACK(tp);
    430  1.35   thorpej 
    431   1.1       cgd 	if (tp->t_template)
    432  1.26   thorpej 		FREE(tp->t_template, M_MBUF);
    433  1.57   thorpej 	pool_put(&tcpcb_pool, tp);
    434   1.1       cgd 	inp->inp_ppcb = 0;
    435   1.1       cgd 	soisdisconnected(so);
    436   1.1       cgd 	in_pcbdetach(inp);
    437   1.1       cgd 	tcpstat.tcps_closed++;
    438   1.1       cgd 	return ((struct tcpcb *)0);
    439   1.1       cgd }
    440   1.1       cgd 
    441  1.35   thorpej int
    442  1.35   thorpej tcp_freeq(tp)
    443  1.35   thorpej 	struct tcpcb *tp;
    444  1.35   thorpej {
    445  1.35   thorpej 	register struct ipqent *qe;
    446  1.35   thorpej 	int rv = 0;
    447  1.49      matt #ifdef TCPREASS_DEBUG
    448  1.49      matt 	int i = 0;
    449  1.49      matt #endif
    450  1.35   thorpej 
    451  1.35   thorpej 	while ((qe = tp->segq.lh_first) != NULL) {
    452  1.49      matt #ifdef TCPREASS_DEBUG
    453  1.49      matt 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
    454  1.49      matt 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
    455  1.49      matt 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
    456  1.49      matt #endif
    457  1.35   thorpej 		LIST_REMOVE(qe, ipqe_q);
    458  1.49      matt 		LIST_REMOVE(qe, ipqe_timeq);
    459  1.35   thorpej 		m_freem(qe->ipqe_m);
    460  1.35   thorpej 		FREE(qe, M_IPQ);
    461  1.35   thorpej 		rv = 1;
    462  1.35   thorpej 	}
    463  1.35   thorpej 	return (rv);
    464  1.35   thorpej }
    465  1.35   thorpej 
    466  1.35   thorpej /*
    467  1.35   thorpej  * Protocol drain routine.  Called when memory is in short supply.
    468  1.35   thorpej  */
    469   1.7   mycroft void
    470   1.1       cgd tcp_drain()
    471   1.1       cgd {
    472  1.35   thorpej 	register struct inpcb *inp;
    473  1.35   thorpej 	register struct tcpcb *tp;
    474   1.1       cgd 
    475  1.35   thorpej 	/*
    476  1.35   thorpej 	 * Free the sequence queue of all TCP connections.
    477  1.35   thorpej 	 */
    478  1.35   thorpej 	inp = tcbtable.inpt_queue.cqh_first;
    479  1.35   thorpej 	if (inp)						/* XXX */
    480  1.35   thorpej 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
    481  1.35   thorpej 	    inp = inp->inp_queue.cqe_next) {
    482  1.35   thorpej 		if ((tp = intotcpcb(inp)) != NULL) {
    483  1.35   thorpej 			if (tcp_freeq(tp))
    484  1.35   thorpej 				tcpstat.tcps_connsdrained++;
    485  1.35   thorpej 		}
    486  1.35   thorpej 	}
    487   1.1       cgd }
    488   1.1       cgd 
    489   1.1       cgd /*
    490   1.1       cgd  * Notify a tcp user of an asynchronous error;
    491   1.1       cgd  * store error as soft error, but wake up user
    492   1.1       cgd  * (for now, won't do anything until can select for soft error).
    493   1.1       cgd  */
    494   1.7   mycroft void
    495   1.1       cgd tcp_notify(inp, error)
    496  1.10   mycroft 	struct inpcb *inp;
    497   1.1       cgd 	int error;
    498   1.1       cgd {
    499  1.10   mycroft 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    500  1.10   mycroft 	register struct socket *so = inp->inp_socket;
    501   1.1       cgd 
    502  1.10   mycroft 	/*
    503  1.10   mycroft 	 * Ignore some errors if we are hooked up.
    504  1.10   mycroft 	 * If connection hasn't completed, has retransmitted several times,
    505  1.10   mycroft 	 * and receives a second error, give up now.  This is better
    506  1.10   mycroft 	 * than waiting a long time to establish a connection that
    507  1.10   mycroft 	 * can never complete.
    508  1.10   mycroft 	 */
    509  1.10   mycroft 	if (tp->t_state == TCPS_ESTABLISHED &&
    510  1.10   mycroft 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    511  1.10   mycroft 	      error == EHOSTDOWN)) {
    512  1.10   mycroft 		return;
    513  1.12   mycroft 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    514  1.12   mycroft 	    tp->t_rxtshift > 3 && tp->t_softerror)
    515  1.10   mycroft 		so->so_error = error;
    516  1.10   mycroft 	else
    517  1.10   mycroft 		tp->t_softerror = error;
    518  1.10   mycroft 	wakeup((caddr_t) &so->so_timeo);
    519  1.10   mycroft 	sorwakeup(so);
    520  1.10   mycroft 	sowwakeup(so);
    521   1.1       cgd }
    522   1.1       cgd 
    523  1.22  christos void *
    524  1.22  christos tcp_ctlinput(cmd, sa, v)
    525   1.1       cgd 	int cmd;
    526   1.1       cgd 	struct sockaddr *sa;
    527  1.22  christos 	register void *v;
    528   1.1       cgd {
    529  1.22  christos 	register struct ip *ip = v;
    530   1.1       cgd 	register struct tcphdr *th;
    531  1.19   mycroft 	extern int inetctlerrmap[];
    532   1.7   mycroft 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    533  1.19   mycroft 	int errno;
    534  1.27   thorpej 	int nmatch;
    535   1.1       cgd 
    536  1.18   mycroft 	if ((unsigned)cmd >= PRC_NCMDS)
    537  1.22  christos 		return NULL;
    538  1.18   mycroft 	errno = inetctlerrmap[cmd];
    539  1.17   mycroft 	if (cmd == PRC_QUENCH)
    540  1.17   mycroft 		notify = tcp_quench;
    541  1.17   mycroft 	else if (PRC_IS_REDIRECT(cmd))
    542  1.17   mycroft 		notify = in_rtchange, ip = 0;
    543  1.32       kml 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
    544  1.31       kml 		notify = tcp_mtudisc, ip = 0;
    545  1.16   mycroft 	else if (cmd == PRC_HOSTDEAD)
    546  1.17   mycroft 		ip = 0;
    547  1.18   mycroft 	else if (errno == 0)
    548  1.22  christos 		return NULL;
    549  1.17   mycroft 	if (ip) {
    550   1.1       cgd 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    551  1.27   thorpej 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    552  1.27   thorpej 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    553  1.27   thorpej 		if (nmatch == 0 && syn_cache_count &&
    554  1.27   thorpej 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    555  1.27   thorpej 		    inetctlerrmap[cmd] == ENETUNREACH ||
    556  1.27   thorpej 		    inetctlerrmap[cmd] == EHOSTDOWN))
    557  1.27   thorpej 			syn_cache_unreach(ip, th);
    558  1.17   mycroft 	} else
    559  1.27   thorpej 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    560  1.23   mycroft 		    notify);
    561  1.22  christos 	return NULL;
    562   1.1       cgd }
    563   1.1       cgd 
    564   1.1       cgd /*
    565  1.55   thorpej  * When a source quence is received, we are being notifed of congestion.
    566  1.55   thorpej  * Close the congestion window down to the Loss Window (one segment).
    567  1.55   thorpej  * We will gradually open it again as we proceed.
    568   1.1       cgd  */
    569   1.7   mycroft void
    570   1.7   mycroft tcp_quench(inp, errno)
    571   1.1       cgd 	struct inpcb *inp;
    572   1.7   mycroft 	int errno;
    573   1.1       cgd {
    574   1.1       cgd 	struct tcpcb *tp = intotcpcb(inp);
    575   1.1       cgd 
    576   1.1       cgd 	if (tp)
    577  1.55   thorpej 		tp->snd_cwnd = tp->t_segsz;
    578  1.28   thorpej }
    579  1.31       kml 
    580  1.31       kml /*
    581  1.31       kml  * On receipt of path MTU corrections, flush old route and replace it
    582  1.31       kml  * with the new one.  Retransmit all unacknowledged packets, to ensure
    583  1.31       kml  * that all packets will be received.
    584  1.31       kml  */
    585  1.31       kml void
    586  1.31       kml tcp_mtudisc(inp, errno)
    587  1.31       kml 	struct inpcb *inp;
    588  1.31       kml 	int errno;
    589  1.31       kml {
    590  1.31       kml 	struct tcpcb *tp = intotcpcb(inp);
    591  1.31       kml 	struct rtentry *rt = in_pcbrtentry(inp);
    592  1.31       kml 
    593  1.31       kml 	if (tp != 0) {
    594  1.31       kml 		if (rt != 0) {
    595  1.36   thorpej 			/*
    596  1.36   thorpej 			 * If this was not a host route, remove and realloc.
    597  1.36   thorpej 			 */
    598  1.31       kml 			if ((rt->rt_flags & RTF_HOST) == 0) {
    599  1.31       kml 				in_rtchange(inp, errno);
    600  1.31       kml 				if ((rt = in_pcbrtentry(inp)) == 0)
    601  1.31       kml 					return;
    602  1.31       kml 			}
    603  1.36   thorpej 
    604  1.36   thorpej 			/*
    605  1.36   thorpej 			 * Slow start out of the error condition.  We
    606  1.36   thorpej 			 * use the MTU because we know it's smaller
    607  1.36   thorpej 			 * than the previously transmitted segment.
    608  1.56   thorpej 			 *
    609  1.56   thorpej 			 * Note: This is more conservative than the
    610  1.56   thorpej 			 * suggestion in draft-floyd-incr-init-win-03.
    611  1.36   thorpej 			 */
    612  1.33       kml 			if (rt->rt_rmx.rmx_mtu != 0)
    613  1.36   thorpej 				tp->snd_cwnd =
    614  1.46   thorpej 				    TCP_INITIAL_WINDOW(tcp_init_win,
    615  1.46   thorpej 				    rt->rt_rmx.rmx_mtu);
    616  1.31       kml 		}
    617  1.31       kml 
    618  1.36   thorpej 		/*
    619  1.36   thorpej 		 * Resend unacknowledged packets.
    620  1.36   thorpej 		 */
    621  1.31       kml 		tp->snd_nxt = tp->snd_una;
    622  1.31       kml 		tcp_output(tp);
    623  1.31       kml 	}
    624  1.31       kml }
    625  1.31       kml 
    626  1.28   thorpej 
    627  1.28   thorpej /*
    628  1.28   thorpej  * Compute the MSS to advertise to the peer.  Called only during
    629  1.28   thorpej  * the 3-way handshake.  If we are the server (peer initiated
    630  1.53       kml  * connection), we are called with a pointer to the interface
    631  1.53       kml  * on which the SYN packet arrived.  If we are the client (we
    632  1.53       kml  * initiated connection), we are called with a pointer to the
    633  1.53       kml  * interface out which this connection should go.
    634  1.28   thorpej  */
    635  1.47       kml u_long
    636  1.47       kml tcp_mss_to_advertise(ifp)
    637  1.47       kml 	const struct ifnet *ifp;
    638  1.28   thorpej {
    639  1.28   thorpej 	extern u_long in_maxmtu;
    640  1.47       kml 	u_long mss = 0;
    641  1.28   thorpej 
    642  1.28   thorpej 	/*
    643  1.28   thorpej 	 * In order to avoid defeating path MTU discovery on the peer,
    644  1.28   thorpej 	 * we advertise the max MTU of all attached networks as our MSS,
    645  1.28   thorpej 	 * per RFC 1191, section 3.1.
    646  1.47       kml 	 *
    647  1.47       kml 	 * We provide the option to advertise just the MTU of
    648  1.47       kml 	 * the interface on which we hope this connection will
    649  1.47       kml 	 * be receiving.  If we are responding to a SYN, we
    650  1.47       kml 	 * will have a pretty good idea about this, but when
    651  1.47       kml 	 * initiating a connection there is a bit more doubt.
    652  1.47       kml 	 *
    653  1.47       kml 	 * We also need to ensure that loopback has a large enough
    654  1.47       kml 	 * MSS, as the loopback MTU is never included in in_maxmtu.
    655  1.28   thorpej 	 */
    656  1.28   thorpej 
    657  1.47       kml 	if (ifp != NULL)
    658  1.47       kml 		mss = ifp->if_mtu;
    659  1.47       kml 
    660  1.47       kml 	if (tcp_mss_ifmtu == 0)
    661  1.47       kml 		mss = max(in_maxmtu, mss);
    662  1.47       kml 
    663  1.47       kml 	if (mss > sizeof(struct tcpiphdr))
    664  1.47       kml 		mss -= sizeof(struct tcpiphdr);
    665  1.47       kml 
    666  1.47       kml 	mss = max(tcp_mssdflt, mss);
    667  1.28   thorpej 	return (mss);
    668  1.28   thorpej }
    669  1.28   thorpej 
    670  1.28   thorpej /*
    671  1.28   thorpej  * Set connection variables based on the peer's advertised MSS.
    672  1.28   thorpej  * We are passed the TCPCB for the actual connection.  If we
    673  1.28   thorpej  * are the server, we are called by the compressed state engine
    674  1.28   thorpej  * when the 3-way handshake is complete.  If we are the client,
    675  1.28   thorpej  * we are called when we recieve the SYN,ACK from the server.
    676  1.28   thorpej  *
    677  1.28   thorpej  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    678  1.28   thorpej  * before this routine is called!
    679  1.28   thorpej  */
    680  1.28   thorpej void
    681  1.28   thorpej tcp_mss_from_peer(tp, offer)
    682  1.28   thorpej 	struct tcpcb *tp;
    683  1.28   thorpej 	int offer;
    684  1.28   thorpej {
    685  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    686  1.28   thorpej 	struct socket *so = inp->inp_socket;
    687  1.28   thorpej #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    688  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    689  1.28   thorpej #endif
    690  1.28   thorpej 	u_long bufsize;
    691  1.28   thorpej 	int mss;
    692  1.28   thorpej 
    693  1.28   thorpej 	/*
    694  1.42       kml 	 * As per RFC1122, use the default MSS value, unless they
    695  1.42       kml 	 * sent us an offer.  Do not accept offers less than 32 bytes.
    696  1.28   thorpej 	 */
    697  1.42       kml 	mss = tcp_mssdflt;
    698  1.28   thorpej 	if (offer)
    699  1.28   thorpej 		mss = offer;
    700  1.28   thorpej 	mss = max(mss, 32);		/* sanity */
    701  1.54       kml 	tp->t_peermss = mss;
    702  1.44       kml 	mss -= (tcp_optlen(tp) + ip_optlen(tp->t_inpcb));
    703  1.28   thorpej 
    704  1.28   thorpej 	/*
    705  1.28   thorpej 	 * If there's a pipesize, change the socket buffer to that size.
    706  1.28   thorpej 	 * Make the socket buffer an integral number of MSS units.  If
    707  1.28   thorpej 	 * the MSS is larger than the socket buffer, artificially decrease
    708  1.28   thorpej 	 * the MSS.
    709  1.28   thorpej 	 */
    710  1.28   thorpej #ifdef RTV_SPIPE
    711  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    712  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_sendpipe;
    713  1.28   thorpej 	else
    714  1.28   thorpej #endif
    715  1.28   thorpej 		bufsize = so->so_snd.sb_hiwat;
    716  1.28   thorpej 	if (bufsize < mss)
    717  1.28   thorpej 		mss = bufsize;
    718  1.28   thorpej 	else {
    719  1.28   thorpej 		bufsize = roundup(bufsize, mss);
    720  1.28   thorpej 		if (bufsize > sb_max)
    721  1.28   thorpej 			bufsize = sb_max;
    722  1.28   thorpej 		(void) sbreserve(&so->so_snd, bufsize);
    723  1.28   thorpej 	}
    724  1.33       kml 	tp->t_segsz = mss;
    725  1.28   thorpej 
    726  1.28   thorpej #ifdef RTV_SSTHRESH
    727  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    728  1.28   thorpej 		/*
    729  1.28   thorpej 		 * There's some sort of gateway or interface buffer
    730  1.28   thorpej 		 * limit on the path.  Use this to set the slow
    731  1.28   thorpej 		 * start threshold, but set the threshold to no less
    732  1.28   thorpej 		 * than 2 * MSS.
    733  1.28   thorpej 		 */
    734  1.28   thorpej 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    735  1.28   thorpej 	}
    736  1.28   thorpej #endif
    737  1.28   thorpej }
    738  1.28   thorpej 
    739  1.28   thorpej /*
    740  1.28   thorpej  * Processing necessary when a TCP connection is established.
    741  1.28   thorpej  */
    742  1.28   thorpej void
    743  1.28   thorpej tcp_established(tp)
    744  1.28   thorpej 	struct tcpcb *tp;
    745  1.28   thorpej {
    746  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    747  1.28   thorpej 	struct socket *so = inp->inp_socket;
    748  1.28   thorpej #ifdef RTV_RPIPE
    749  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    750  1.28   thorpej #endif
    751  1.28   thorpej 	u_long bufsize;
    752  1.28   thorpej 
    753  1.28   thorpej 	tp->t_state = TCPS_ESTABLISHED;
    754  1.51   thorpej 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
    755  1.28   thorpej 
    756  1.28   thorpej #ifdef RTV_RPIPE
    757  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    758  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_recvpipe;
    759  1.28   thorpej 	else
    760  1.28   thorpej #endif
    761  1.28   thorpej 		bufsize = so->so_rcv.sb_hiwat;
    762  1.28   thorpej 	if (bufsize > tp->t_ourmss) {
    763  1.28   thorpej 		bufsize = roundup(bufsize, tp->t_ourmss);
    764  1.28   thorpej 		if (bufsize > sb_max)
    765  1.28   thorpej 			bufsize = sb_max;
    766  1.28   thorpej 		(void) sbreserve(&so->so_rcv, bufsize);
    767  1.28   thorpej 	}
    768  1.28   thorpej }
    769  1.28   thorpej 
    770  1.28   thorpej /*
    771  1.28   thorpej  * Check if there's an initial rtt or rttvar.  Convert from the
    772  1.28   thorpej  * route-table units to scaled multiples of the slow timeout timer.
    773  1.28   thorpej  * Called only during the 3-way handshake.
    774  1.28   thorpej  */
    775  1.28   thorpej void
    776  1.28   thorpej tcp_rmx_rtt(tp)
    777  1.28   thorpej 	struct tcpcb *tp;
    778  1.28   thorpej {
    779  1.28   thorpej #ifdef RTV_RTT
    780  1.28   thorpej 	struct rtentry *rt;
    781  1.28   thorpej 	int rtt;
    782  1.28   thorpej 
    783  1.28   thorpej 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    784  1.28   thorpej 		return;
    785  1.28   thorpej 
    786  1.28   thorpej 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    787  1.28   thorpej 		/*
    788  1.28   thorpej 		 * XXX The lock bit for MTU indicates that the value
    789  1.28   thorpej 		 * is also a minimum value; this is subject to time.
    790  1.28   thorpej 		 */
    791  1.28   thorpej 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    792  1.43       kml 			TCPT_RANGESET(tp->t_rttmin,
    793  1.43       kml 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
    794  1.43       kml 			    TCPTV_MIN, TCPTV_REXMTMAX);
    795  1.28   thorpej 		tp->t_srtt = rtt /
    796  1.28   thorpej 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    797  1.28   thorpej 		if (rt->rt_rmx.rmx_rttvar) {
    798  1.28   thorpej 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    799  1.28   thorpej 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    800  1.28   thorpej 				(TCP_RTTVAR_SHIFT + 2));
    801  1.28   thorpej 		} else {
    802  1.28   thorpej 			/* Default variation is +- 1 rtt */
    803  1.28   thorpej 			tp->t_rttvar =
    804  1.28   thorpej 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    805  1.28   thorpej 		}
    806  1.28   thorpej 		TCPT_RANGESET(tp->t_rxtcur,
    807  1.28   thorpej 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    808  1.28   thorpej 		    tp->t_rttmin, TCPTV_REXMTMAX);
    809  1.28   thorpej 	}
    810  1.28   thorpej #endif
    811  1.29  explorer }
    812  1.29  explorer 
    813  1.30  explorer tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
    814  1.30  explorer 
    815  1.29  explorer /*
    816  1.29  explorer  * Get a new sequence value given a tcp control block
    817  1.29  explorer  */
    818  1.29  explorer tcp_seq
    819  1.29  explorer tcp_new_iss(tp, len, addin)
    820  1.29  explorer 	void            *tp;
    821  1.29  explorer 	u_long           len;
    822  1.29  explorer 	tcp_seq		 addin;
    823  1.29  explorer {
    824  1.29  explorer 	tcp_seq          tcp_iss;
    825  1.29  explorer 
    826  1.29  explorer 	/*
    827  1.29  explorer 	 * add randomness about this connection, but do not estimate
    828  1.29  explorer 	 * entropy from the timing, since the physical device driver would
    829  1.29  explorer 	 * have done that for us.
    830  1.29  explorer 	 */
    831  1.30  explorer #if NRND > 0
    832  1.29  explorer 	if (tp != NULL)
    833  1.29  explorer 		rnd_add_data(NULL, tp, len, 0);
    834  1.30  explorer #endif
    835  1.29  explorer 
    836  1.29  explorer 	/*
    837  1.29  explorer 	 * randomize.
    838  1.29  explorer 	 */
    839  1.30  explorer #if NRND > 0
    840  1.29  explorer 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    841  1.30  explorer #else
    842  1.30  explorer 	tcp_iss = random();
    843  1.30  explorer #endif
    844  1.29  explorer 
    845  1.29  explorer 	/*
    846  1.29  explorer 	 * If we were asked to add some amount to a known value,
    847  1.29  explorer 	 * we will take a random value obtained above, mask off the upper
    848  1.29  explorer 	 * bits, and add in the known value.  We also add in a constant to
    849  1.29  explorer 	 * ensure that we are at least a certain distance from the original
    850  1.29  explorer 	 * value.
    851  1.29  explorer 	 *
    852  1.29  explorer 	 * This is used when an old connection is in timed wait
    853  1.29  explorer 	 * and we have a new one coming in, for instance.
    854  1.29  explorer 	 */
    855  1.29  explorer 	if (addin != 0) {
    856  1.29  explorer #ifdef TCPISS_DEBUG
    857  1.29  explorer 		printf("Random %08x, ", tcp_iss);
    858  1.29  explorer #endif
    859  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    860  1.30  explorer 		tcp_iss = tcp_iss + addin + TCP_ISSINCR;
    861  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    862  1.30  explorer 		tcp_iss += tcp_iss_seq;
    863  1.29  explorer #ifdef TCPISS_DEBUG
    864  1.29  explorer 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    865  1.29  explorer #endif
    866  1.29  explorer 	} else {
    867  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    868  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    869  1.29  explorer 		tcp_iss += tcp_iss_seq;
    870  1.29  explorer #ifdef TCPISS_DEBUG
    871  1.29  explorer 		printf("ISS %08x\n", tcp_iss);
    872  1.29  explorer #endif
    873  1.29  explorer 	}
    874  1.29  explorer 
    875  1.48   thorpej 	if (tcp_compat_42) {
    876  1.48   thorpej 		/*
    877  1.48   thorpej 		 * Limit it to the positive range for really old TCP
    878  1.48   thorpej 		 * implementations.
    879  1.48   thorpej 		 */
    880  1.48   thorpej 		if ((int)tcp_iss < 0)
    881  1.48   thorpej 			tcp_iss &= 0x7fffffff;		/* XXX */
    882  1.48   thorpej 	}
    883  1.29  explorer 
    884  1.29  explorer 	return tcp_iss;
    885   1.1       cgd }
    886  1.42       kml 
    887  1.42       kml 
    888  1.42       kml /*
    889  1.42       kml  * Determine the length of the TCP options for this connection.
    890  1.42       kml  *
    891  1.42       kml  * XXX:  What do we do for SACK, when we add that?  Just reserve
    892  1.42       kml  *       all of the space?  Otherwise we can't exactly be incrementing
    893  1.42       kml  *       cwnd by an amount that varies depending on the amount we last
    894  1.42       kml  *       had to SACK!
    895  1.42       kml  */
    896  1.42       kml 
    897  1.42       kml u_int
    898  1.42       kml tcp_optlen(tp)
    899  1.42       kml 	struct tcpcb *tp;
    900  1.42       kml {
    901  1.42       kml 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
    902  1.42       kml 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
    903  1.42       kml 		return TCPOLEN_TSTAMP_APPA;
    904  1.42       kml 	else
    905  1.42       kml 		return 0;
    906  1.42       kml }
    907