Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.30
      1  1.30  explorer /*	$NetBSD: tcp_subr.c,v 1.30 1997/10/13 00:48:12 explorer Exp $	*/
      2  1.11       cgd 
      3   1.1       cgd /*
      4  1.10   mycroft  * Copyright (c) 1982, 1986, 1988, 1990, 1993
      5  1.10   mycroft  *	The Regents of the University of California.  All rights reserved.
      6   1.1       cgd  *
      7   1.1       cgd  * Redistribution and use in source and binary forms, with or without
      8   1.1       cgd  * modification, are permitted provided that the following conditions
      9   1.1       cgd  * are met:
     10   1.1       cgd  * 1. Redistributions of source code must retain the above copyright
     11   1.1       cgd  *    notice, this list of conditions and the following disclaimer.
     12   1.1       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     13   1.1       cgd  *    notice, this list of conditions and the following disclaimer in the
     14   1.1       cgd  *    documentation and/or other materials provided with the distribution.
     15   1.1       cgd  * 3. All advertising materials mentioning features or use of this software
     16   1.1       cgd  *    must display the following acknowledgement:
     17   1.1       cgd  *	This product includes software developed by the University of
     18   1.1       cgd  *	California, Berkeley and its contributors.
     19   1.1       cgd  * 4. Neither the name of the University nor the names of its contributors
     20   1.1       cgd  *    may be used to endorse or promote products derived from this software
     21   1.1       cgd  *    without specific prior written permission.
     22   1.1       cgd  *
     23   1.1       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24   1.1       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25   1.1       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26   1.1       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27   1.1       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28   1.1       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29   1.1       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30   1.1       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31   1.1       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32   1.1       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33   1.1       cgd  * SUCH DAMAGE.
     34   1.1       cgd  *
     35  1.11       cgd  *	@(#)tcp_subr.c	8.1 (Berkeley) 6/10/93
     36   1.1       cgd  */
     37   1.1       cgd 
     38  1.30  explorer #include "rnd.h"
     39  1.30  explorer 
     40   1.5   mycroft #include <sys/param.h>
     41  1.10   mycroft #include <sys/proc.h>
     42   1.5   mycroft #include <sys/systm.h>
     43   1.5   mycroft #include <sys/malloc.h>
     44   1.5   mycroft #include <sys/mbuf.h>
     45   1.5   mycroft #include <sys/socket.h>
     46   1.5   mycroft #include <sys/socketvar.h>
     47   1.5   mycroft #include <sys/protosw.h>
     48   1.5   mycroft #include <sys/errno.h>
     49  1.27   thorpej #include <sys/kernel.h>
     50  1.30  explorer #if NRND > 0
     51  1.29  explorer #include <sys/rnd.h>
     52  1.30  explorer #endif
     53   1.1       cgd 
     54   1.5   mycroft #include <net/route.h>
     55   1.5   mycroft #include <net/if.h>
     56   1.1       cgd 
     57   1.5   mycroft #include <netinet/in.h>
     58   1.5   mycroft #include <netinet/in_systm.h>
     59   1.5   mycroft #include <netinet/ip.h>
     60   1.5   mycroft #include <netinet/in_pcb.h>
     61   1.5   mycroft #include <netinet/ip_var.h>
     62   1.5   mycroft #include <netinet/ip_icmp.h>
     63   1.5   mycroft #include <netinet/tcp.h>
     64   1.5   mycroft #include <netinet/tcp_fsm.h>
     65   1.5   mycroft #include <netinet/tcp_seq.h>
     66   1.5   mycroft #include <netinet/tcp_timer.h>
     67   1.5   mycroft #include <netinet/tcp_var.h>
     68   1.5   mycroft #include <netinet/tcpip.h>
     69   1.1       cgd 
     70   1.1       cgd /* patchable/settable parameters for tcp */
     71   1.1       cgd int 	tcp_mssdflt = TCP_MSS;
     72   1.1       cgd int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
     73  1.10   mycroft int	tcp_do_rfc1323 = 1;
     74   1.1       cgd 
     75  1.21   mycroft #ifndef TCBHASHSIZE
     76  1.21   mycroft #define	TCBHASHSIZE	128
     77  1.21   mycroft #endif
     78  1.21   mycroft int	tcbhashsize = TCBHASHSIZE;
     79   1.1       cgd 
     80   1.1       cgd /*
     81   1.1       cgd  * Tcp initialization
     82   1.1       cgd  */
     83   1.7   mycroft void
     84   1.1       cgd tcp_init()
     85   1.1       cgd {
     86   1.1       cgd 
     87  1.24   mycroft 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
     88   1.1       cgd 	if (max_protohdr < sizeof(struct tcpiphdr))
     89   1.1       cgd 		max_protohdr = sizeof(struct tcpiphdr);
     90   1.1       cgd 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
     91   1.1       cgd 		panic("tcp_init");
     92   1.1       cgd }
     93   1.1       cgd 
     94   1.1       cgd /*
     95   1.1       cgd  * Create template to be used to send tcp packets on a connection.
     96   1.1       cgd  * Call after host entry created, allocates an mbuf and fills
     97   1.1       cgd  * in a skeletal tcp/ip header, minimizing the amount of work
     98   1.1       cgd  * necessary when the connection is used.
     99   1.1       cgd  */
    100   1.1       cgd struct tcpiphdr *
    101   1.1       cgd tcp_template(tp)
    102   1.1       cgd 	struct tcpcb *tp;
    103   1.1       cgd {
    104   1.1       cgd 	register struct inpcb *inp = tp->t_inpcb;
    105   1.1       cgd 	register struct tcpiphdr *n;
    106   1.1       cgd 
    107   1.1       cgd 	if ((n = tp->t_template) == 0) {
    108  1.26   thorpej 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    109  1.26   thorpej 		    M_MBUF, M_NOWAIT);
    110  1.26   thorpej 		if (n == NULL)
    111   1.1       cgd 			return (0);
    112   1.1       cgd 	}
    113  1.20       cgd 	bzero(n->ti_x1, sizeof n->ti_x1);
    114   1.1       cgd 	n->ti_pr = IPPROTO_TCP;
    115   1.1       cgd 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    116   1.1       cgd 	n->ti_src = inp->inp_laddr;
    117   1.1       cgd 	n->ti_dst = inp->inp_faddr;
    118   1.1       cgd 	n->ti_sport = inp->inp_lport;
    119   1.1       cgd 	n->ti_dport = inp->inp_fport;
    120   1.1       cgd 	n->ti_seq = 0;
    121   1.1       cgd 	n->ti_ack = 0;
    122   1.1       cgd 	n->ti_x2 = 0;
    123   1.1       cgd 	n->ti_off = 5;
    124   1.1       cgd 	n->ti_flags = 0;
    125   1.1       cgd 	n->ti_win = 0;
    126   1.1       cgd 	n->ti_sum = 0;
    127   1.1       cgd 	n->ti_urp = 0;
    128   1.1       cgd 	return (n);
    129   1.1       cgd }
    130   1.1       cgd 
    131   1.1       cgd /*
    132   1.1       cgd  * Send a single message to the TCP at address specified by
    133   1.1       cgd  * the given TCP/IP header.  If m == 0, then we make a copy
    134   1.1       cgd  * of the tcpiphdr at ti and send directly to the addressed host.
    135   1.1       cgd  * This is used to force keep alive messages out using the TCP
    136   1.1       cgd  * template for a connection tp->t_template.  If flags are given
    137   1.1       cgd  * then we send a message back to the TCP which originated the
    138   1.1       cgd  * segment ti, and discard the mbuf containing it and any other
    139   1.1       cgd  * attached mbufs.
    140   1.1       cgd  *
    141   1.1       cgd  * In any case the ack and sequence number of the transmitted
    142   1.1       cgd  * segment are as specified by the parameters.
    143   1.1       cgd  */
    144  1.27   thorpej int
    145   1.1       cgd tcp_respond(tp, ti, m, ack, seq, flags)
    146   1.1       cgd 	struct tcpcb *tp;
    147   1.1       cgd 	register struct tcpiphdr *ti;
    148   1.1       cgd 	register struct mbuf *m;
    149   1.1       cgd 	tcp_seq ack, seq;
    150   1.1       cgd 	int flags;
    151   1.1       cgd {
    152   1.1       cgd 	register int tlen;
    153   1.1       cgd 	int win = 0;
    154   1.1       cgd 	struct route *ro = 0;
    155   1.1       cgd 
    156   1.1       cgd 	if (tp) {
    157   1.1       cgd 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    158   1.1       cgd 		ro = &tp->t_inpcb->inp_route;
    159   1.1       cgd 	}
    160   1.1       cgd 	if (m == 0) {
    161   1.1       cgd 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    162   1.1       cgd 		if (m == NULL)
    163  1.27   thorpej 			return (ENOBUFS);
    164   1.1       cgd #ifdef TCP_COMPAT_42
    165   1.1       cgd 		tlen = 1;
    166   1.1       cgd #else
    167   1.1       cgd 		tlen = 0;
    168   1.1       cgd #endif
    169   1.1       cgd 		m->m_data += max_linkhdr;
    170   1.1       cgd 		*mtod(m, struct tcpiphdr *) = *ti;
    171   1.1       cgd 		ti = mtod(m, struct tcpiphdr *);
    172   1.1       cgd 		flags = TH_ACK;
    173   1.1       cgd 	} else {
    174   1.1       cgd 		m_freem(m->m_next);
    175   1.1       cgd 		m->m_next = 0;
    176   1.1       cgd 		m->m_data = (caddr_t)ti;
    177   1.1       cgd 		m->m_len = sizeof (struct tcpiphdr);
    178   1.1       cgd 		tlen = 0;
    179  1.10   mycroft #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    180  1.13       cgd 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    181  1.13       cgd 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    182   1.1       cgd #undef xchg
    183   1.1       cgd 	}
    184  1.20       cgd 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    185   1.1       cgd 	ti->ti_seq = htonl(seq);
    186   1.1       cgd 	ti->ti_ack = htonl(ack);
    187   1.1       cgd 	ti->ti_x2 = 0;
    188  1.27   thorpej 	if ((flags & TH_SYN) == 0) {
    189  1.27   thorpej 		if (tp)
    190  1.27   thorpej 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    191  1.27   thorpej 		else
    192  1.27   thorpej 			ti->ti_win = htons((u_int16_t)win);
    193  1.27   thorpej 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    194  1.27   thorpej 		tlen += sizeof (struct tcphdr);
    195  1.27   thorpej 	} else
    196  1.27   thorpej 		tlen += ti->ti_off << 2;
    197  1.27   thorpej 	ti->ti_len = htons((u_int16_t)tlen);
    198  1.27   thorpej 	tlen += sizeof (struct ip);
    199  1.27   thorpej 	m->m_len = tlen;
    200  1.27   thorpej 	m->m_pkthdr.len = tlen;
    201  1.27   thorpej 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    202   1.1       cgd 	ti->ti_flags = flags;
    203   1.1       cgd 	ti->ti_urp = 0;
    204  1.10   mycroft 	ti->ti_sum = 0;
    205   1.1       cgd 	ti->ti_sum = in_cksum(m, tlen);
    206   1.1       cgd 	((struct ip *)ti)->ip_len = tlen;
    207  1.10   mycroft 	((struct ip *)ti)->ip_ttl = ip_defttl;
    208  1.27   thorpej 	return ip_output(m, NULL, ro, 0, NULL);
    209   1.1       cgd }
    210   1.1       cgd 
    211   1.1       cgd /*
    212   1.1       cgd  * Create a new TCP control block, making an
    213   1.1       cgd  * empty reassembly queue and hooking it to the argument
    214   1.1       cgd  * protocol control block.
    215   1.1       cgd  */
    216   1.1       cgd struct tcpcb *
    217   1.1       cgd tcp_newtcpcb(inp)
    218   1.1       cgd 	struct inpcb *inp;
    219   1.1       cgd {
    220   1.1       cgd 	register struct tcpcb *tp;
    221   1.1       cgd 
    222  1.10   mycroft 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
    223  1.10   mycroft 	if (tp == NULL)
    224   1.1       cgd 		return ((struct tcpcb *)0);
    225  1.23   mycroft 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    226  1.20       cgd 	LIST_INIT(&tp->segq);
    227   1.1       cgd 	tp->t_maxseg = tcp_mssdflt;
    228  1.28   thorpej 	tp->t_ourmss = tcp_mssdflt;
    229   1.1       cgd 
    230  1.10   mycroft 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
    231   1.1       cgd 	tp->t_inpcb = inp;
    232   1.1       cgd 	/*
    233   1.1       cgd 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    234   1.1       cgd 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    235   1.1       cgd 	 * reasonable initial retransmit time.
    236   1.1       cgd 	 */
    237   1.1       cgd 	tp->t_srtt = TCPTV_SRTTBASE;
    238  1.15   mycroft 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    239   1.1       cgd 	tp->t_rttmin = TCPTV_MIN;
    240  1.15   mycroft 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    241   1.1       cgd 	    TCPTV_MIN, TCPTV_REXMTMAX);
    242  1.10   mycroft 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    243  1.10   mycroft 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    244  1.10   mycroft 	inp->inp_ip.ip_ttl = ip_defttl;
    245   1.1       cgd 	inp->inp_ppcb = (caddr_t)tp;
    246   1.1       cgd 	return (tp);
    247   1.1       cgd }
    248   1.1       cgd 
    249   1.1       cgd /*
    250   1.1       cgd  * Drop a TCP connection, reporting
    251   1.1       cgd  * the specified error.  If connection is synchronized,
    252   1.1       cgd  * then send a RST to peer.
    253   1.1       cgd  */
    254   1.1       cgd struct tcpcb *
    255   1.1       cgd tcp_drop(tp, errno)
    256   1.1       cgd 	register struct tcpcb *tp;
    257   1.1       cgd 	int errno;
    258   1.1       cgd {
    259   1.1       cgd 	struct socket *so = tp->t_inpcb->inp_socket;
    260   1.1       cgd 
    261   1.1       cgd 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    262   1.1       cgd 		tp->t_state = TCPS_CLOSED;
    263   1.1       cgd 		(void) tcp_output(tp);
    264   1.1       cgd 		tcpstat.tcps_drops++;
    265   1.1       cgd 	} else
    266   1.1       cgd 		tcpstat.tcps_conndrops++;
    267   1.1       cgd 	if (errno == ETIMEDOUT && tp->t_softerror)
    268   1.1       cgd 		errno = tp->t_softerror;
    269   1.1       cgd 	so->so_error = errno;
    270   1.1       cgd 	return (tcp_close(tp));
    271   1.1       cgd }
    272   1.1       cgd 
    273   1.1       cgd /*
    274   1.1       cgd  * Close a TCP control block:
    275   1.1       cgd  *	discard all space held by the tcp
    276   1.1       cgd  *	discard internet protocol block
    277   1.1       cgd  *	wake up any sleepers
    278   1.1       cgd  */
    279   1.1       cgd struct tcpcb *
    280   1.1       cgd tcp_close(tp)
    281   1.1       cgd 	register struct tcpcb *tp;
    282   1.1       cgd {
    283  1.20       cgd 	register struct ipqent *qe;
    284   1.1       cgd 	struct inpcb *inp = tp->t_inpcb;
    285   1.1       cgd 	struct socket *so = inp->inp_socket;
    286   1.1       cgd #ifdef RTV_RTT
    287   1.1       cgd 	register struct rtentry *rt;
    288   1.1       cgd 
    289   1.1       cgd 	/*
    290   1.1       cgd 	 * If we sent enough data to get some meaningful characteristics,
    291  1.10   mycroft 	 * save them in the routing entry.  'Enough' is arbitrarily
    292   1.1       cgd 	 * defined as the sendpipesize (default 4K) * 16.  This would
    293   1.1       cgd 	 * give us 16 rtt samples assuming we only get one sample per
    294   1.1       cgd 	 * window (the usual case on a long haul net).  16 samples is
    295   1.1       cgd 	 * enough for the srtt filter to converge to within 5% of the correct
    296   1.1       cgd 	 * value; fewer samples and we could save a very bogus rtt.
    297   1.1       cgd 	 *
    298   1.1       cgd 	 * Don't update the default route's characteristics and don't
    299   1.1       cgd 	 * update anything that the user "locked".
    300   1.1       cgd 	 */
    301   1.1       cgd 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    302   1.1       cgd 	    (rt = inp->inp_route.ro_rt) &&
    303  1.23   mycroft 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    304  1.22  christos 		register u_long i = 0;
    305   1.1       cgd 
    306   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    307   1.1       cgd 			i = tp->t_srtt *
    308  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    309   1.1       cgd 			if (rt->rt_rmx.rmx_rtt && i)
    310   1.1       cgd 				/*
    311   1.1       cgd 				 * filter this update to half the old & half
    312   1.1       cgd 				 * the new values, converting scale.
    313   1.1       cgd 				 * See route.h and tcp_var.h for a
    314   1.1       cgd 				 * description of the scaling constants.
    315   1.1       cgd 				 */
    316   1.1       cgd 				rt->rt_rmx.rmx_rtt =
    317   1.1       cgd 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    318   1.1       cgd 			else
    319   1.1       cgd 				rt->rt_rmx.rmx_rtt = i;
    320   1.1       cgd 		}
    321   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    322   1.1       cgd 			i = tp->t_rttvar *
    323  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    324   1.1       cgd 			if (rt->rt_rmx.rmx_rttvar && i)
    325   1.1       cgd 				rt->rt_rmx.rmx_rttvar =
    326   1.1       cgd 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    327   1.1       cgd 			else
    328   1.1       cgd 				rt->rt_rmx.rmx_rttvar = i;
    329   1.1       cgd 		}
    330   1.1       cgd 		/*
    331   1.1       cgd 		 * update the pipelimit (ssthresh) if it has been updated
    332   1.1       cgd 		 * already or if a pipesize was specified & the threshhold
    333   1.1       cgd 		 * got below half the pipesize.  I.e., wait for bad news
    334   1.1       cgd 		 * before we start updating, then update on both good
    335   1.1       cgd 		 * and bad news.
    336   1.1       cgd 		 */
    337  1.22  christos 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    338  1.22  christos 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    339   1.1       cgd 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    340   1.1       cgd 			/*
    341   1.1       cgd 			 * convert the limit from user data bytes to
    342   1.1       cgd 			 * packets then to packet data bytes.
    343   1.1       cgd 			 */
    344   1.1       cgd 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
    345   1.1       cgd 			if (i < 2)
    346   1.1       cgd 				i = 2;
    347   1.1       cgd 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
    348   1.1       cgd 			if (rt->rt_rmx.rmx_ssthresh)
    349   1.1       cgd 				rt->rt_rmx.rmx_ssthresh =
    350   1.1       cgd 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    351   1.1       cgd 			else
    352   1.1       cgd 				rt->rt_rmx.rmx_ssthresh = i;
    353   1.1       cgd 		}
    354   1.1       cgd 	}
    355   1.9   mycroft #endif /* RTV_RTT */
    356   1.1       cgd 	/* free the reassembly queue, if any */
    357  1.20       cgd 	while ((qe = tp->segq.lh_first) != NULL) {
    358  1.20       cgd 		LIST_REMOVE(qe, ipqe_q);
    359  1.20       cgd 		m_freem(qe->ipqe_m);
    360  1.20       cgd 		FREE(qe, M_IPQ);
    361   1.1       cgd 	}
    362   1.1       cgd 	if (tp->t_template)
    363  1.26   thorpej 		FREE(tp->t_template, M_MBUF);
    364  1.10   mycroft 	free(tp, M_PCB);
    365   1.1       cgd 	inp->inp_ppcb = 0;
    366   1.1       cgd 	soisdisconnected(so);
    367   1.1       cgd 	in_pcbdetach(inp);
    368   1.1       cgd 	tcpstat.tcps_closed++;
    369   1.1       cgd 	return ((struct tcpcb *)0);
    370   1.1       cgd }
    371   1.1       cgd 
    372   1.7   mycroft void
    373   1.1       cgd tcp_drain()
    374   1.1       cgd {
    375   1.1       cgd 
    376   1.1       cgd }
    377   1.1       cgd 
    378   1.1       cgd /*
    379   1.1       cgd  * Notify a tcp user of an asynchronous error;
    380   1.1       cgd  * store error as soft error, but wake up user
    381   1.1       cgd  * (for now, won't do anything until can select for soft error).
    382   1.1       cgd  */
    383   1.7   mycroft void
    384   1.1       cgd tcp_notify(inp, error)
    385  1.10   mycroft 	struct inpcb *inp;
    386   1.1       cgd 	int error;
    387   1.1       cgd {
    388  1.10   mycroft 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    389  1.10   mycroft 	register struct socket *so = inp->inp_socket;
    390   1.1       cgd 
    391  1.10   mycroft 	/*
    392  1.10   mycroft 	 * Ignore some errors if we are hooked up.
    393  1.10   mycroft 	 * If connection hasn't completed, has retransmitted several times,
    394  1.10   mycroft 	 * and receives a second error, give up now.  This is better
    395  1.10   mycroft 	 * than waiting a long time to establish a connection that
    396  1.10   mycroft 	 * can never complete.
    397  1.10   mycroft 	 */
    398  1.10   mycroft 	if (tp->t_state == TCPS_ESTABLISHED &&
    399  1.10   mycroft 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    400  1.10   mycroft 	      error == EHOSTDOWN)) {
    401  1.10   mycroft 		return;
    402  1.12   mycroft 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    403  1.12   mycroft 	    tp->t_rxtshift > 3 && tp->t_softerror)
    404  1.10   mycroft 		so->so_error = error;
    405  1.10   mycroft 	else
    406  1.10   mycroft 		tp->t_softerror = error;
    407  1.10   mycroft 	wakeup((caddr_t) &so->so_timeo);
    408  1.10   mycroft 	sorwakeup(so);
    409  1.10   mycroft 	sowwakeup(so);
    410   1.1       cgd }
    411   1.1       cgd 
    412  1.22  christos void *
    413  1.22  christos tcp_ctlinput(cmd, sa, v)
    414   1.1       cgd 	int cmd;
    415   1.1       cgd 	struct sockaddr *sa;
    416  1.22  christos 	register void *v;
    417   1.1       cgd {
    418  1.22  christos 	register struct ip *ip = v;
    419   1.1       cgd 	register struct tcphdr *th;
    420  1.19   mycroft 	extern int inetctlerrmap[];
    421   1.7   mycroft 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    422  1.19   mycroft 	int errno;
    423  1.27   thorpej 	int nmatch;
    424   1.1       cgd 
    425  1.18   mycroft 	if ((unsigned)cmd >= PRC_NCMDS)
    426  1.22  christos 		return NULL;
    427  1.18   mycroft 	errno = inetctlerrmap[cmd];
    428  1.17   mycroft 	if (cmd == PRC_QUENCH)
    429  1.17   mycroft 		notify = tcp_quench;
    430  1.17   mycroft 	else if (PRC_IS_REDIRECT(cmd))
    431  1.17   mycroft 		notify = in_rtchange, ip = 0;
    432  1.16   mycroft 	else if (cmd == PRC_HOSTDEAD)
    433  1.17   mycroft 		ip = 0;
    434  1.18   mycroft 	else if (errno == 0)
    435  1.22  christos 		return NULL;
    436  1.17   mycroft 	if (ip) {
    437   1.1       cgd 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    438  1.27   thorpej 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    439  1.27   thorpej 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    440  1.27   thorpej 		if (nmatch == 0 && syn_cache_count &&
    441  1.27   thorpej 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    442  1.27   thorpej 		    inetctlerrmap[cmd] == ENETUNREACH ||
    443  1.27   thorpej 		    inetctlerrmap[cmd] == EHOSTDOWN))
    444  1.27   thorpej 			syn_cache_unreach(ip, th);
    445  1.17   mycroft 	} else
    446  1.27   thorpej 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    447  1.23   mycroft 		    notify);
    448  1.22  christos 	return NULL;
    449   1.1       cgd }
    450   1.1       cgd 
    451   1.1       cgd /*
    452   1.1       cgd  * When a source quench is received, close congestion window
    453   1.1       cgd  * to one segment.  We will gradually open it again as we proceed.
    454   1.1       cgd  */
    455   1.7   mycroft void
    456   1.7   mycroft tcp_quench(inp, errno)
    457   1.1       cgd 	struct inpcb *inp;
    458   1.7   mycroft 	int errno;
    459   1.1       cgd {
    460   1.1       cgd 	struct tcpcb *tp = intotcpcb(inp);
    461   1.1       cgd 
    462   1.1       cgd 	if (tp)
    463   1.1       cgd 		tp->snd_cwnd = tp->t_maxseg;
    464  1.28   thorpej }
    465  1.28   thorpej 
    466  1.28   thorpej /*
    467  1.28   thorpej  * Compute the MSS to advertise to the peer.  Called only during
    468  1.28   thorpej  * the 3-way handshake.  If we are the server (peer initiated
    469  1.28   thorpej  * connection), we are called with the TCPCB for the listen
    470  1.28   thorpej  * socket.  If we are the client (we initiated connection), we
    471  1.28   thorpej  * are called witht he TCPCB for the actual connection.
    472  1.28   thorpej  */
    473  1.28   thorpej int
    474  1.28   thorpej tcp_mss_to_advertise(tp)
    475  1.28   thorpej 	const struct tcpcb *tp;
    476  1.28   thorpej {
    477  1.28   thorpej 	extern u_long in_maxmtu;
    478  1.28   thorpej 	struct inpcb *inp;
    479  1.28   thorpej 	struct socket *so;
    480  1.28   thorpej 	int mss;
    481  1.28   thorpej 
    482  1.28   thorpej 	inp = tp->t_inpcb;
    483  1.28   thorpej 	so = inp->inp_socket;
    484  1.28   thorpej 
    485  1.28   thorpej 	/*
    486  1.28   thorpej 	 * In order to avoid defeating path MTU discovery on the peer,
    487  1.28   thorpej 	 * we advertise the max MTU of all attached networks as our MSS,
    488  1.28   thorpej 	 * per RFC 1191, section 3.1.
    489  1.28   thorpej 	 *
    490  1.28   thorpej 	 * XXX Should we allow room for the timestamp option if
    491  1.28   thorpej 	 * XXX rfc1323 is enabled?
    492  1.28   thorpej 	 */
    493  1.28   thorpej 	mss = in_maxmtu - sizeof(struct tcpiphdr);
    494  1.28   thorpej 
    495  1.28   thorpej 	return (mss);
    496  1.28   thorpej }
    497  1.28   thorpej 
    498  1.28   thorpej /*
    499  1.28   thorpej  * Set connection variables based on the peer's advertised MSS.
    500  1.28   thorpej  * We are passed the TCPCB for the actual connection.  If we
    501  1.28   thorpej  * are the server, we are called by the compressed state engine
    502  1.28   thorpej  * when the 3-way handshake is complete.  If we are the client,
    503  1.28   thorpej  * we are called when we recieve the SYN,ACK from the server.
    504  1.28   thorpej  *
    505  1.28   thorpej  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    506  1.28   thorpej  * before this routine is called!
    507  1.28   thorpej  */
    508  1.28   thorpej void
    509  1.28   thorpej tcp_mss_from_peer(tp, offer)
    510  1.28   thorpej 	struct tcpcb *tp;
    511  1.28   thorpej 	int offer;
    512  1.28   thorpej {
    513  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    514  1.28   thorpej 	struct socket *so = inp->inp_socket;
    515  1.28   thorpej #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    516  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    517  1.28   thorpej #endif
    518  1.28   thorpej 	u_long bufsize;
    519  1.28   thorpej 	int mss;
    520  1.28   thorpej 
    521  1.28   thorpej 	/*
    522  1.28   thorpej 	 * Assume our MSS is the MSS of the peer, unless they sent us
    523  1.28   thorpej 	 * an offer.  Do not accept offers less than 32 bytes.
    524  1.28   thorpej 	 */
    525  1.28   thorpej 	mss = tp->t_ourmss;
    526  1.28   thorpej 	if (offer)
    527  1.28   thorpej 		mss = offer;
    528  1.28   thorpej 	mss = max(mss, 32);		/* sanity */
    529  1.28   thorpej 
    530  1.28   thorpej 	/*
    531  1.28   thorpej 	 * If there's a pipesize, change the socket buffer to that size.
    532  1.28   thorpej 	 * Make the socket buffer an integral number of MSS units.  If
    533  1.28   thorpej 	 * the MSS is larger than the socket buffer, artificially decrease
    534  1.28   thorpej 	 * the MSS.
    535  1.28   thorpej 	 */
    536  1.28   thorpej #ifdef RTV_SPIPE
    537  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    538  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_sendpipe;
    539  1.28   thorpej 	else
    540  1.28   thorpej #endif
    541  1.28   thorpej 		bufsize = so->so_snd.sb_hiwat;
    542  1.28   thorpej 	if (bufsize < mss)
    543  1.28   thorpej 		mss = bufsize;
    544  1.28   thorpej 	else {
    545  1.28   thorpej 		bufsize = roundup(bufsize, mss);
    546  1.28   thorpej 		if (bufsize > sb_max)
    547  1.28   thorpej 			bufsize = sb_max;
    548  1.28   thorpej 		(void) sbreserve(&so->so_snd, bufsize);
    549  1.28   thorpej 	}
    550  1.28   thorpej 	tp->t_maxseg = mss;
    551  1.28   thorpej 
    552  1.28   thorpej 	/* Initialize the initial congestion window. */
    553  1.28   thorpej 	tp->snd_cwnd = mss;
    554  1.28   thorpej 
    555  1.28   thorpej #ifdef RTV_SSTHRESH
    556  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    557  1.28   thorpej 		/*
    558  1.28   thorpej 		 * There's some sort of gateway or interface buffer
    559  1.28   thorpej 		 * limit on the path.  Use this to set the slow
    560  1.28   thorpej 		 * start threshold, but set the threshold to no less
    561  1.28   thorpej 		 * than 2 * MSS.
    562  1.28   thorpej 		 */
    563  1.28   thorpej 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    564  1.28   thorpej 	}
    565  1.28   thorpej #endif
    566  1.28   thorpej }
    567  1.28   thorpej 
    568  1.28   thorpej /*
    569  1.28   thorpej  * Processing necessary when a TCP connection is established.
    570  1.28   thorpej  */
    571  1.28   thorpej void
    572  1.28   thorpej tcp_established(tp)
    573  1.28   thorpej 	struct tcpcb *tp;
    574  1.28   thorpej {
    575  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    576  1.28   thorpej 	struct socket *so = inp->inp_socket;
    577  1.28   thorpej #ifdef RTV_RPIPE
    578  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    579  1.28   thorpej #endif
    580  1.28   thorpej 	u_long bufsize;
    581  1.28   thorpej 
    582  1.28   thorpej 	tp->t_state = TCPS_ESTABLISHED;
    583  1.28   thorpej 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    584  1.28   thorpej 
    585  1.28   thorpej #ifdef RTV_RPIPE
    586  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    587  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_recvpipe;
    588  1.28   thorpej 	else
    589  1.28   thorpej #endif
    590  1.28   thorpej 		bufsize = so->so_rcv.sb_hiwat;
    591  1.28   thorpej 	if (bufsize > tp->t_ourmss) {
    592  1.28   thorpej 		bufsize = roundup(bufsize, tp->t_ourmss);
    593  1.28   thorpej 		if (bufsize > sb_max)
    594  1.28   thorpej 			bufsize = sb_max;
    595  1.28   thorpej 		(void) sbreserve(&so->so_rcv, bufsize);
    596  1.28   thorpej 	}
    597  1.28   thorpej }
    598  1.28   thorpej 
    599  1.28   thorpej /*
    600  1.28   thorpej  * Check if there's an initial rtt or rttvar.  Convert from the
    601  1.28   thorpej  * route-table units to scaled multiples of the slow timeout timer.
    602  1.28   thorpej  * Called only during the 3-way handshake.
    603  1.28   thorpej  */
    604  1.28   thorpej void
    605  1.28   thorpej tcp_rmx_rtt(tp)
    606  1.28   thorpej 	struct tcpcb *tp;
    607  1.28   thorpej {
    608  1.28   thorpej #ifdef RTV_RTT
    609  1.28   thorpej 	struct rtentry *rt;
    610  1.28   thorpej 	int rtt;
    611  1.28   thorpej 
    612  1.28   thorpej 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    613  1.28   thorpej 		return;
    614  1.28   thorpej 
    615  1.28   thorpej 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    616  1.28   thorpej 		/*
    617  1.28   thorpej 		 * XXX The lock bit for MTU indicates that the value
    618  1.28   thorpej 		 * is also a minimum value; this is subject to time.
    619  1.28   thorpej 		 */
    620  1.28   thorpej 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    621  1.28   thorpej 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
    622  1.28   thorpej 		tp->t_srtt = rtt /
    623  1.28   thorpej 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    624  1.28   thorpej 		if (rt->rt_rmx.rmx_rttvar) {
    625  1.28   thorpej 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    626  1.28   thorpej 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    627  1.28   thorpej 				(TCP_RTTVAR_SHIFT + 2));
    628  1.28   thorpej 		} else {
    629  1.28   thorpej 			/* Default variation is +- 1 rtt */
    630  1.28   thorpej 			tp->t_rttvar =
    631  1.28   thorpej 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    632  1.28   thorpej 		}
    633  1.28   thorpej 		TCPT_RANGESET(tp->t_rxtcur,
    634  1.28   thorpej 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    635  1.28   thorpej 		    tp->t_rttmin, TCPTV_REXMTMAX);
    636  1.28   thorpej 	}
    637  1.28   thorpej #endif
    638  1.29  explorer }
    639  1.29  explorer 
    640  1.30  explorer tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
    641  1.30  explorer 
    642  1.29  explorer /*
    643  1.29  explorer  * Get a new sequence value given a tcp control block
    644  1.29  explorer  */
    645  1.29  explorer tcp_seq
    646  1.29  explorer tcp_new_iss(tp, len, addin)
    647  1.29  explorer 	void            *tp;
    648  1.29  explorer 	u_long           len;
    649  1.29  explorer 	tcp_seq		 addin;
    650  1.29  explorer {
    651  1.29  explorer 	tcp_seq          tcp_iss;
    652  1.29  explorer 
    653  1.29  explorer 	/*
    654  1.29  explorer 	 * add randomness about this connection, but do not estimate
    655  1.29  explorer 	 * entropy from the timing, since the physical device driver would
    656  1.29  explorer 	 * have done that for us.
    657  1.29  explorer 	 */
    658  1.30  explorer #if NRND > 0
    659  1.29  explorer 	if (tp != NULL)
    660  1.29  explorer 		rnd_add_data(NULL, tp, len, 0);
    661  1.30  explorer #endif
    662  1.29  explorer 
    663  1.29  explorer 	/*
    664  1.29  explorer 	 * randomize.
    665  1.29  explorer 	 */
    666  1.30  explorer #if NRND > 0
    667  1.29  explorer 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    668  1.30  explorer #else
    669  1.30  explorer 	tcp_iss = random();
    670  1.30  explorer #endif
    671  1.29  explorer 
    672  1.29  explorer 	/*
    673  1.29  explorer 	 * If we were asked to add some amount to a known value,
    674  1.29  explorer 	 * we will take a random value obtained above, mask off the upper
    675  1.29  explorer 	 * bits, and add in the known value.  We also add in a constant to
    676  1.29  explorer 	 * ensure that we are at least a certain distance from the original
    677  1.29  explorer 	 * value.
    678  1.29  explorer 	 *
    679  1.29  explorer 	 * This is used when an old connection is in timed wait
    680  1.29  explorer 	 * and we have a new one coming in, for instance.
    681  1.29  explorer 	 */
    682  1.29  explorer 	if (addin != 0) {
    683  1.29  explorer #ifdef TCPISS_DEBUG
    684  1.29  explorer 		printf("Random %08x, ", tcp_iss);
    685  1.29  explorer #endif
    686  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    687  1.30  explorer 		tcp_iss = tcp_iss + addin + TCP_ISSINCR;
    688  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    689  1.30  explorer 		tcp_iss += tcp_iss_seq;
    690  1.29  explorer #ifdef TCPISS_DEBUG
    691  1.29  explorer 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    692  1.29  explorer #endif
    693  1.29  explorer 	} else {
    694  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    695  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    696  1.29  explorer 		tcp_iss += tcp_iss_seq;
    697  1.29  explorer #ifdef TCPISS_DEBUG
    698  1.29  explorer 		printf("ISS %08x\n", tcp_iss);
    699  1.29  explorer #endif
    700  1.29  explorer 	}
    701  1.29  explorer 
    702  1.29  explorer #ifdef TCP_COMPAT_42
    703  1.29  explorer 	/*
    704  1.29  explorer 	 * limit it to the positive range for really old TCP implementations
    705  1.29  explorer 	 */
    706  1.29  explorer 	if ((int)tcp_iss < 0)
    707  1.29  explorer 		tcp_iss &= 0x7fffffff;		/* XXX */
    708  1.29  explorer #endif
    709  1.29  explorer 
    710  1.29  explorer 	return tcp_iss;
    711   1.1       cgd }
    712