Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.37
      1  1.37   thorpej /*	$NetBSD: tcp_subr.c,v 1.37 1997/12/31 03:31:26 thorpej 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  1.36   thorpej int	tcp_init_win = 1;
     75   1.1       cgd 
     76  1.21   mycroft #ifndef TCBHASHSIZE
     77  1.21   mycroft #define	TCBHASHSIZE	128
     78  1.21   mycroft #endif
     79  1.21   mycroft int	tcbhashsize = TCBHASHSIZE;
     80   1.1       cgd 
     81  1.35   thorpej int	tcp_freeq __P((struct tcpcb *));
     82  1.35   thorpej 
     83   1.1       cgd /*
     84   1.1       cgd  * Tcp initialization
     85   1.1       cgd  */
     86   1.7   mycroft void
     87   1.1       cgd tcp_init()
     88   1.1       cgd {
     89   1.1       cgd 
     90  1.24   mycroft 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
     91  1.37   thorpej 	LIST_INIT(&tcp_delacks);
     92   1.1       cgd 	if (max_protohdr < sizeof(struct tcpiphdr))
     93   1.1       cgd 		max_protohdr = sizeof(struct tcpiphdr);
     94   1.1       cgd 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
     95   1.1       cgd 		panic("tcp_init");
     96   1.1       cgd }
     97   1.1       cgd 
     98   1.1       cgd /*
     99   1.1       cgd  * Create template to be used to send tcp packets on a connection.
    100   1.1       cgd  * Call after host entry created, allocates an mbuf and fills
    101   1.1       cgd  * in a skeletal tcp/ip header, minimizing the amount of work
    102   1.1       cgd  * necessary when the connection is used.
    103   1.1       cgd  */
    104   1.1       cgd struct tcpiphdr *
    105   1.1       cgd tcp_template(tp)
    106   1.1       cgd 	struct tcpcb *tp;
    107   1.1       cgd {
    108   1.1       cgd 	register struct inpcb *inp = tp->t_inpcb;
    109   1.1       cgd 	register struct tcpiphdr *n;
    110   1.1       cgd 
    111   1.1       cgd 	if ((n = tp->t_template) == 0) {
    112  1.26   thorpej 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    113  1.26   thorpej 		    M_MBUF, M_NOWAIT);
    114  1.26   thorpej 		if (n == NULL)
    115   1.1       cgd 			return (0);
    116   1.1       cgd 	}
    117  1.20       cgd 	bzero(n->ti_x1, sizeof n->ti_x1);
    118   1.1       cgd 	n->ti_pr = IPPROTO_TCP;
    119   1.1       cgd 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    120   1.1       cgd 	n->ti_src = inp->inp_laddr;
    121   1.1       cgd 	n->ti_dst = inp->inp_faddr;
    122   1.1       cgd 	n->ti_sport = inp->inp_lport;
    123   1.1       cgd 	n->ti_dport = inp->inp_fport;
    124   1.1       cgd 	n->ti_seq = 0;
    125   1.1       cgd 	n->ti_ack = 0;
    126   1.1       cgd 	n->ti_x2 = 0;
    127   1.1       cgd 	n->ti_off = 5;
    128   1.1       cgd 	n->ti_flags = 0;
    129   1.1       cgd 	n->ti_win = 0;
    130   1.1       cgd 	n->ti_sum = 0;
    131   1.1       cgd 	n->ti_urp = 0;
    132   1.1       cgd 	return (n);
    133   1.1       cgd }
    134   1.1       cgd 
    135   1.1       cgd /*
    136   1.1       cgd  * Send a single message to the TCP at address specified by
    137   1.1       cgd  * the given TCP/IP header.  If m == 0, then we make a copy
    138   1.1       cgd  * of the tcpiphdr at ti and send directly to the addressed host.
    139   1.1       cgd  * This is used to force keep alive messages out using the TCP
    140   1.1       cgd  * template for a connection tp->t_template.  If flags are given
    141   1.1       cgd  * then we send a message back to the TCP which originated the
    142   1.1       cgd  * segment ti, and discard the mbuf containing it and any other
    143   1.1       cgd  * attached mbufs.
    144   1.1       cgd  *
    145   1.1       cgd  * In any case the ack and sequence number of the transmitted
    146   1.1       cgd  * segment are as specified by the parameters.
    147   1.1       cgd  */
    148  1.27   thorpej int
    149   1.1       cgd tcp_respond(tp, ti, m, ack, seq, flags)
    150   1.1       cgd 	struct tcpcb *tp;
    151   1.1       cgd 	register struct tcpiphdr *ti;
    152   1.1       cgd 	register struct mbuf *m;
    153   1.1       cgd 	tcp_seq ack, seq;
    154   1.1       cgd 	int flags;
    155   1.1       cgd {
    156   1.1       cgd 	register int tlen;
    157   1.1       cgd 	int win = 0;
    158   1.1       cgd 	struct route *ro = 0;
    159   1.1       cgd 
    160   1.1       cgd 	if (tp) {
    161   1.1       cgd 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    162   1.1       cgd 		ro = &tp->t_inpcb->inp_route;
    163   1.1       cgd 	}
    164   1.1       cgd 	if (m == 0) {
    165   1.1       cgd 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    166   1.1       cgd 		if (m == NULL)
    167  1.27   thorpej 			return (ENOBUFS);
    168   1.1       cgd #ifdef TCP_COMPAT_42
    169   1.1       cgd 		tlen = 1;
    170   1.1       cgd #else
    171   1.1       cgd 		tlen = 0;
    172   1.1       cgd #endif
    173   1.1       cgd 		m->m_data += max_linkhdr;
    174   1.1       cgd 		*mtod(m, struct tcpiphdr *) = *ti;
    175   1.1       cgd 		ti = mtod(m, struct tcpiphdr *);
    176   1.1       cgd 		flags = TH_ACK;
    177   1.1       cgd 	} else {
    178   1.1       cgd 		m_freem(m->m_next);
    179   1.1       cgd 		m->m_next = 0;
    180   1.1       cgd 		m->m_data = (caddr_t)ti;
    181   1.1       cgd 		m->m_len = sizeof (struct tcpiphdr);
    182   1.1       cgd 		tlen = 0;
    183  1.10   mycroft #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    184  1.13       cgd 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    185  1.13       cgd 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    186   1.1       cgd #undef xchg
    187   1.1       cgd 	}
    188  1.20       cgd 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    189   1.1       cgd 	ti->ti_seq = htonl(seq);
    190   1.1       cgd 	ti->ti_ack = htonl(ack);
    191   1.1       cgd 	ti->ti_x2 = 0;
    192  1.27   thorpej 	if ((flags & TH_SYN) == 0) {
    193  1.27   thorpej 		if (tp)
    194  1.27   thorpej 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    195  1.27   thorpej 		else
    196  1.27   thorpej 			ti->ti_win = htons((u_int16_t)win);
    197  1.27   thorpej 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    198  1.27   thorpej 		tlen += sizeof (struct tcphdr);
    199  1.27   thorpej 	} else
    200  1.27   thorpej 		tlen += ti->ti_off << 2;
    201  1.27   thorpej 	ti->ti_len = htons((u_int16_t)tlen);
    202  1.27   thorpej 	tlen += sizeof (struct ip);
    203  1.27   thorpej 	m->m_len = tlen;
    204  1.27   thorpej 	m->m_pkthdr.len = tlen;
    205  1.27   thorpej 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    206   1.1       cgd 	ti->ti_flags = flags;
    207   1.1       cgd 	ti->ti_urp = 0;
    208  1.10   mycroft 	ti->ti_sum = 0;
    209   1.1       cgd 	ti->ti_sum = in_cksum(m, tlen);
    210   1.1       cgd 	((struct ip *)ti)->ip_len = tlen;
    211  1.10   mycroft 	((struct ip *)ti)->ip_ttl = ip_defttl;
    212  1.27   thorpej 	return ip_output(m, NULL, ro, 0, NULL);
    213   1.1       cgd }
    214   1.1       cgd 
    215   1.1       cgd /*
    216   1.1       cgd  * Create a new TCP control block, making an
    217   1.1       cgd  * empty reassembly queue and hooking it to the argument
    218   1.1       cgd  * protocol control block.
    219   1.1       cgd  */
    220   1.1       cgd struct tcpcb *
    221   1.1       cgd tcp_newtcpcb(inp)
    222   1.1       cgd 	struct inpcb *inp;
    223   1.1       cgd {
    224   1.1       cgd 	register struct tcpcb *tp;
    225   1.1       cgd 
    226  1.10   mycroft 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
    227  1.10   mycroft 	if (tp == NULL)
    228   1.1       cgd 		return ((struct tcpcb *)0);
    229  1.23   mycroft 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    230  1.20       cgd 	LIST_INIT(&tp->segq);
    231  1.33       kml 	tp->t_peermss = tcp_mssdflt;
    232  1.28   thorpej 	tp->t_ourmss = tcp_mssdflt;
    233  1.33       kml 	tp->t_segsz = tcp_mssdflt;
    234   1.1       cgd 
    235  1.10   mycroft 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
    236   1.1       cgd 	tp->t_inpcb = inp;
    237   1.1       cgd 	/*
    238   1.1       cgd 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    239   1.1       cgd 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    240   1.1       cgd 	 * reasonable initial retransmit time.
    241   1.1       cgd 	 */
    242   1.1       cgd 	tp->t_srtt = TCPTV_SRTTBASE;
    243  1.15   mycroft 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    244   1.1       cgd 	tp->t_rttmin = TCPTV_MIN;
    245  1.15   mycroft 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    246   1.1       cgd 	    TCPTV_MIN, TCPTV_REXMTMAX);
    247  1.10   mycroft 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    248  1.10   mycroft 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    249  1.10   mycroft 	inp->inp_ip.ip_ttl = ip_defttl;
    250   1.1       cgd 	inp->inp_ppcb = (caddr_t)tp;
    251   1.1       cgd 	return (tp);
    252   1.1       cgd }
    253   1.1       cgd 
    254   1.1       cgd /*
    255   1.1       cgd  * Drop a TCP connection, reporting
    256   1.1       cgd  * the specified error.  If connection is synchronized,
    257   1.1       cgd  * then send a RST to peer.
    258   1.1       cgd  */
    259   1.1       cgd struct tcpcb *
    260   1.1       cgd tcp_drop(tp, errno)
    261   1.1       cgd 	register struct tcpcb *tp;
    262   1.1       cgd 	int errno;
    263   1.1       cgd {
    264   1.1       cgd 	struct socket *so = tp->t_inpcb->inp_socket;
    265   1.1       cgd 
    266   1.1       cgd 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    267   1.1       cgd 		tp->t_state = TCPS_CLOSED;
    268   1.1       cgd 		(void) tcp_output(tp);
    269   1.1       cgd 		tcpstat.tcps_drops++;
    270   1.1       cgd 	} else
    271   1.1       cgd 		tcpstat.tcps_conndrops++;
    272   1.1       cgd 	if (errno == ETIMEDOUT && tp->t_softerror)
    273   1.1       cgd 		errno = tp->t_softerror;
    274   1.1       cgd 	so->so_error = errno;
    275   1.1       cgd 	return (tcp_close(tp));
    276   1.1       cgd }
    277   1.1       cgd 
    278   1.1       cgd /*
    279   1.1       cgd  * Close a TCP control block:
    280   1.1       cgd  *	discard all space held by the tcp
    281   1.1       cgd  *	discard internet protocol block
    282   1.1       cgd  *	wake up any sleepers
    283   1.1       cgd  */
    284   1.1       cgd struct tcpcb *
    285   1.1       cgd tcp_close(tp)
    286   1.1       cgd 	register struct tcpcb *tp;
    287   1.1       cgd {
    288   1.1       cgd 	struct inpcb *inp = tp->t_inpcb;
    289   1.1       cgd 	struct socket *so = inp->inp_socket;
    290   1.1       cgd #ifdef RTV_RTT
    291   1.1       cgd 	register struct rtentry *rt;
    292   1.1       cgd 
    293   1.1       cgd 	/*
    294   1.1       cgd 	 * If we sent enough data to get some meaningful characteristics,
    295  1.10   mycroft 	 * save them in the routing entry.  'Enough' is arbitrarily
    296   1.1       cgd 	 * defined as the sendpipesize (default 4K) * 16.  This would
    297   1.1       cgd 	 * give us 16 rtt samples assuming we only get one sample per
    298   1.1       cgd 	 * window (the usual case on a long haul net).  16 samples is
    299   1.1       cgd 	 * enough for the srtt filter to converge to within 5% of the correct
    300   1.1       cgd 	 * value; fewer samples and we could save a very bogus rtt.
    301   1.1       cgd 	 *
    302   1.1       cgd 	 * Don't update the default route's characteristics and don't
    303   1.1       cgd 	 * update anything that the user "locked".
    304   1.1       cgd 	 */
    305   1.1       cgd 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    306   1.1       cgd 	    (rt = inp->inp_route.ro_rt) &&
    307  1.23   mycroft 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    308  1.22  christos 		register u_long i = 0;
    309   1.1       cgd 
    310   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    311   1.1       cgd 			i = tp->t_srtt *
    312  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    313   1.1       cgd 			if (rt->rt_rmx.rmx_rtt && i)
    314   1.1       cgd 				/*
    315   1.1       cgd 				 * filter this update to half the old & half
    316   1.1       cgd 				 * the new values, converting scale.
    317   1.1       cgd 				 * See route.h and tcp_var.h for a
    318   1.1       cgd 				 * description of the scaling constants.
    319   1.1       cgd 				 */
    320   1.1       cgd 				rt->rt_rmx.rmx_rtt =
    321   1.1       cgd 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    322   1.1       cgd 			else
    323   1.1       cgd 				rt->rt_rmx.rmx_rtt = i;
    324   1.1       cgd 		}
    325   1.1       cgd 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    326   1.1       cgd 			i = tp->t_rttvar *
    327  1.25   mycroft 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    328   1.1       cgd 			if (rt->rt_rmx.rmx_rttvar && i)
    329   1.1       cgd 				rt->rt_rmx.rmx_rttvar =
    330   1.1       cgd 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    331   1.1       cgd 			else
    332   1.1       cgd 				rt->rt_rmx.rmx_rttvar = i;
    333   1.1       cgd 		}
    334   1.1       cgd 		/*
    335   1.1       cgd 		 * update the pipelimit (ssthresh) if it has been updated
    336   1.1       cgd 		 * already or if a pipesize was specified & the threshhold
    337   1.1       cgd 		 * got below half the pipesize.  I.e., wait for bad news
    338   1.1       cgd 		 * before we start updating, then update on both good
    339   1.1       cgd 		 * and bad news.
    340   1.1       cgd 		 */
    341  1.22  christos 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    342  1.22  christos 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    343   1.1       cgd 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    344   1.1       cgd 			/*
    345   1.1       cgd 			 * convert the limit from user data bytes to
    346   1.1       cgd 			 * packets then to packet data bytes.
    347   1.1       cgd 			 */
    348  1.33       kml 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
    349   1.1       cgd 			if (i < 2)
    350   1.1       cgd 				i = 2;
    351  1.33       kml 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
    352   1.1       cgd 			if (rt->rt_rmx.rmx_ssthresh)
    353   1.1       cgd 				rt->rt_rmx.rmx_ssthresh =
    354   1.1       cgd 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    355   1.1       cgd 			else
    356   1.1       cgd 				rt->rt_rmx.rmx_ssthresh = i;
    357   1.1       cgd 		}
    358   1.1       cgd 	}
    359   1.9   mycroft #endif /* RTV_RTT */
    360   1.1       cgd 	/* free the reassembly queue, if any */
    361  1.35   thorpej 	(void) tcp_freeq(tp);
    362  1.35   thorpej 
    363   1.1       cgd 	if (tp->t_template)
    364  1.26   thorpej 		FREE(tp->t_template, M_MBUF);
    365  1.10   mycroft 	free(tp, M_PCB);
    366   1.1       cgd 	inp->inp_ppcb = 0;
    367   1.1       cgd 	soisdisconnected(so);
    368   1.1       cgd 	in_pcbdetach(inp);
    369   1.1       cgd 	tcpstat.tcps_closed++;
    370   1.1       cgd 	return ((struct tcpcb *)0);
    371   1.1       cgd }
    372   1.1       cgd 
    373  1.35   thorpej int
    374  1.35   thorpej tcp_freeq(tp)
    375  1.35   thorpej 	struct tcpcb *tp;
    376  1.35   thorpej {
    377  1.35   thorpej 	register struct ipqent *qe;
    378  1.35   thorpej 	int rv = 0;
    379  1.35   thorpej 
    380  1.35   thorpej 	while ((qe = tp->segq.lh_first) != NULL) {
    381  1.35   thorpej 		LIST_REMOVE(qe, ipqe_q);
    382  1.35   thorpej 		m_freem(qe->ipqe_m);
    383  1.35   thorpej 		FREE(qe, M_IPQ);
    384  1.35   thorpej 		rv = 1;
    385  1.35   thorpej 	}
    386  1.35   thorpej 	return (rv);
    387  1.35   thorpej }
    388  1.35   thorpej 
    389  1.35   thorpej /*
    390  1.35   thorpej  * Protocol drain routine.  Called when memory is in short supply.
    391  1.35   thorpej  */
    392   1.7   mycroft void
    393   1.1       cgd tcp_drain()
    394   1.1       cgd {
    395  1.35   thorpej 	register struct inpcb *inp;
    396  1.35   thorpej 	register struct tcpcb *tp;
    397   1.1       cgd 
    398  1.35   thorpej 	/*
    399  1.35   thorpej 	 * Free the sequence queue of all TCP connections.
    400  1.35   thorpej 	 */
    401  1.35   thorpej 	inp = tcbtable.inpt_queue.cqh_first;
    402  1.35   thorpej 	if (inp)						/* XXX */
    403  1.35   thorpej 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
    404  1.35   thorpej 	    inp = inp->inp_queue.cqe_next) {
    405  1.35   thorpej 		if ((tp = intotcpcb(inp)) != NULL) {
    406  1.35   thorpej 			if (tcp_freeq(tp))
    407  1.35   thorpej 				tcpstat.tcps_connsdrained++;
    408  1.35   thorpej 		}
    409  1.35   thorpej 	}
    410   1.1       cgd }
    411   1.1       cgd 
    412   1.1       cgd /*
    413   1.1       cgd  * Notify a tcp user of an asynchronous error;
    414   1.1       cgd  * store error as soft error, but wake up user
    415   1.1       cgd  * (for now, won't do anything until can select for soft error).
    416   1.1       cgd  */
    417   1.7   mycroft void
    418   1.1       cgd tcp_notify(inp, error)
    419  1.10   mycroft 	struct inpcb *inp;
    420   1.1       cgd 	int error;
    421   1.1       cgd {
    422  1.10   mycroft 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    423  1.10   mycroft 	register struct socket *so = inp->inp_socket;
    424   1.1       cgd 
    425  1.10   mycroft 	/*
    426  1.10   mycroft 	 * Ignore some errors if we are hooked up.
    427  1.10   mycroft 	 * If connection hasn't completed, has retransmitted several times,
    428  1.10   mycroft 	 * and receives a second error, give up now.  This is better
    429  1.10   mycroft 	 * than waiting a long time to establish a connection that
    430  1.10   mycroft 	 * can never complete.
    431  1.10   mycroft 	 */
    432  1.10   mycroft 	if (tp->t_state == TCPS_ESTABLISHED &&
    433  1.10   mycroft 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    434  1.10   mycroft 	      error == EHOSTDOWN)) {
    435  1.10   mycroft 		return;
    436  1.12   mycroft 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    437  1.12   mycroft 	    tp->t_rxtshift > 3 && tp->t_softerror)
    438  1.10   mycroft 		so->so_error = error;
    439  1.10   mycroft 	else
    440  1.10   mycroft 		tp->t_softerror = error;
    441  1.10   mycroft 	wakeup((caddr_t) &so->so_timeo);
    442  1.10   mycroft 	sorwakeup(so);
    443  1.10   mycroft 	sowwakeup(so);
    444   1.1       cgd }
    445   1.1       cgd 
    446  1.22  christos void *
    447  1.22  christos tcp_ctlinput(cmd, sa, v)
    448   1.1       cgd 	int cmd;
    449   1.1       cgd 	struct sockaddr *sa;
    450  1.22  christos 	register void *v;
    451   1.1       cgd {
    452  1.22  christos 	register struct ip *ip = v;
    453   1.1       cgd 	register struct tcphdr *th;
    454  1.19   mycroft 	extern int inetctlerrmap[];
    455   1.7   mycroft 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    456  1.19   mycroft 	int errno;
    457  1.27   thorpej 	int nmatch;
    458   1.1       cgd 
    459  1.18   mycroft 	if ((unsigned)cmd >= PRC_NCMDS)
    460  1.22  christos 		return NULL;
    461  1.18   mycroft 	errno = inetctlerrmap[cmd];
    462  1.17   mycroft 	if (cmd == PRC_QUENCH)
    463  1.17   mycroft 		notify = tcp_quench;
    464  1.17   mycroft 	else if (PRC_IS_REDIRECT(cmd))
    465  1.17   mycroft 		notify = in_rtchange, ip = 0;
    466  1.32       kml 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
    467  1.31       kml 		notify = tcp_mtudisc, ip = 0;
    468  1.16   mycroft 	else if (cmd == PRC_HOSTDEAD)
    469  1.17   mycroft 		ip = 0;
    470  1.18   mycroft 	else if (errno == 0)
    471  1.22  christos 		return NULL;
    472  1.17   mycroft 	if (ip) {
    473   1.1       cgd 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    474  1.27   thorpej 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    475  1.27   thorpej 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    476  1.27   thorpej 		if (nmatch == 0 && syn_cache_count &&
    477  1.27   thorpej 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    478  1.27   thorpej 		    inetctlerrmap[cmd] == ENETUNREACH ||
    479  1.27   thorpej 		    inetctlerrmap[cmd] == EHOSTDOWN))
    480  1.27   thorpej 			syn_cache_unreach(ip, th);
    481  1.17   mycroft 	} else
    482  1.27   thorpej 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    483  1.23   mycroft 		    notify);
    484  1.22  christos 	return NULL;
    485   1.1       cgd }
    486   1.1       cgd 
    487   1.1       cgd /*
    488   1.1       cgd  * When a source quench is received, close congestion window
    489   1.1       cgd  * to one segment.  We will gradually open it again as we proceed.
    490   1.1       cgd  */
    491   1.7   mycroft void
    492   1.7   mycroft tcp_quench(inp, errno)
    493   1.1       cgd 	struct inpcb *inp;
    494   1.7   mycroft 	int errno;
    495   1.1       cgd {
    496   1.1       cgd 	struct tcpcb *tp = intotcpcb(inp);
    497   1.1       cgd 
    498   1.1       cgd 	if (tp)
    499  1.33       kml 		tp->snd_cwnd = tp->t_segsz;
    500  1.28   thorpej }
    501  1.31       kml 
    502  1.31       kml /*
    503  1.31       kml  * On receipt of path MTU corrections, flush old route and replace it
    504  1.31       kml  * with the new one.  Retransmit all unacknowledged packets, to ensure
    505  1.31       kml  * that all packets will be received.
    506  1.31       kml  */
    507  1.31       kml void
    508  1.31       kml tcp_mtudisc(inp, errno)
    509  1.31       kml 	struct inpcb *inp;
    510  1.31       kml 	int errno;
    511  1.31       kml {
    512  1.31       kml 	struct tcpcb *tp = intotcpcb(inp);
    513  1.31       kml 	struct rtentry *rt = in_pcbrtentry(inp);
    514  1.31       kml 
    515  1.31       kml 	if (tp != 0) {
    516  1.31       kml 		if (rt != 0) {
    517  1.36   thorpej 			/*
    518  1.36   thorpej 			 * If this was not a host route, remove and realloc.
    519  1.36   thorpej 			 */
    520  1.31       kml 			if ((rt->rt_flags & RTF_HOST) == 0) {
    521  1.31       kml 				in_rtchange(inp, errno);
    522  1.31       kml 				if ((rt = in_pcbrtentry(inp)) == 0)
    523  1.31       kml 					return;
    524  1.31       kml 			}
    525  1.36   thorpej 
    526  1.36   thorpej 			/*
    527  1.36   thorpej 			 * Slow start out of the error condition.  We
    528  1.36   thorpej 			 * use the MTU because we know it's smaller
    529  1.36   thorpej 			 * than the previously transmitted segment.
    530  1.36   thorpej 			 */
    531  1.33       kml 			if (rt->rt_rmx.rmx_mtu != 0)
    532  1.36   thorpej 				tp->snd_cwnd =
    533  1.36   thorpej 				    TCP_INITIAL_WINDOW(rt->rt_rmx.rmx_mtu);
    534  1.31       kml 		}
    535  1.31       kml 
    536  1.36   thorpej 		/*
    537  1.36   thorpej 		 * Resend unacknowledged packets.
    538  1.36   thorpej 		 */
    539  1.31       kml 		tp->snd_nxt = tp->snd_una;
    540  1.31       kml 		tcp_output(tp);
    541  1.31       kml 	}
    542  1.31       kml }
    543  1.31       kml 
    544  1.28   thorpej 
    545  1.28   thorpej /*
    546  1.28   thorpej  * Compute the MSS to advertise to the peer.  Called only during
    547  1.28   thorpej  * the 3-way handshake.  If we are the server (peer initiated
    548  1.28   thorpej  * connection), we are called with the TCPCB for the listen
    549  1.28   thorpej  * socket.  If we are the client (we initiated connection), we
    550  1.28   thorpej  * are called witht he TCPCB for the actual connection.
    551  1.28   thorpej  */
    552  1.28   thorpej int
    553  1.28   thorpej tcp_mss_to_advertise(tp)
    554  1.28   thorpej 	const struct tcpcb *tp;
    555  1.28   thorpej {
    556  1.28   thorpej 	extern u_long in_maxmtu;
    557  1.28   thorpej 	struct inpcb *inp;
    558  1.28   thorpej 	struct socket *so;
    559  1.28   thorpej 	int mss;
    560  1.28   thorpej 
    561  1.28   thorpej 	inp = tp->t_inpcb;
    562  1.28   thorpej 	so = inp->inp_socket;
    563  1.28   thorpej 
    564  1.28   thorpej 	/*
    565  1.28   thorpej 	 * In order to avoid defeating path MTU discovery on the peer,
    566  1.28   thorpej 	 * we advertise the max MTU of all attached networks as our MSS,
    567  1.28   thorpej 	 * per RFC 1191, section 3.1.
    568  1.28   thorpej 	 *
    569  1.28   thorpej 	 * XXX Should we allow room for the timestamp option if
    570  1.28   thorpej 	 * XXX rfc1323 is enabled?
    571  1.28   thorpej 	 */
    572  1.28   thorpej 	mss = in_maxmtu - sizeof(struct tcpiphdr);
    573  1.28   thorpej 
    574  1.28   thorpej 	return (mss);
    575  1.28   thorpej }
    576  1.28   thorpej 
    577  1.28   thorpej /*
    578  1.28   thorpej  * Set connection variables based on the peer's advertised MSS.
    579  1.28   thorpej  * We are passed the TCPCB for the actual connection.  If we
    580  1.28   thorpej  * are the server, we are called by the compressed state engine
    581  1.28   thorpej  * when the 3-way handshake is complete.  If we are the client,
    582  1.28   thorpej  * we are called when we recieve the SYN,ACK from the server.
    583  1.28   thorpej  *
    584  1.28   thorpej  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    585  1.28   thorpej  * before this routine is called!
    586  1.28   thorpej  */
    587  1.28   thorpej void
    588  1.28   thorpej tcp_mss_from_peer(tp, offer)
    589  1.28   thorpej 	struct tcpcb *tp;
    590  1.28   thorpej 	int offer;
    591  1.28   thorpej {
    592  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    593  1.28   thorpej 	struct socket *so = inp->inp_socket;
    594  1.28   thorpej #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    595  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    596  1.28   thorpej #endif
    597  1.28   thorpej 	u_long bufsize;
    598  1.28   thorpej 	int mss;
    599  1.28   thorpej 
    600  1.28   thorpej 	/*
    601  1.28   thorpej 	 * Assume our MSS is the MSS of the peer, unless they sent us
    602  1.28   thorpej 	 * an offer.  Do not accept offers less than 32 bytes.
    603  1.28   thorpej 	 */
    604  1.28   thorpej 	mss = tp->t_ourmss;
    605  1.28   thorpej 	if (offer)
    606  1.28   thorpej 		mss = offer;
    607  1.28   thorpej 	mss = max(mss, 32);		/* sanity */
    608  1.28   thorpej 
    609  1.28   thorpej 	/*
    610  1.28   thorpej 	 * If there's a pipesize, change the socket buffer to that size.
    611  1.28   thorpej 	 * Make the socket buffer an integral number of MSS units.  If
    612  1.28   thorpej 	 * the MSS is larger than the socket buffer, artificially decrease
    613  1.28   thorpej 	 * the MSS.
    614  1.28   thorpej 	 */
    615  1.28   thorpej #ifdef RTV_SPIPE
    616  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    617  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_sendpipe;
    618  1.28   thorpej 	else
    619  1.28   thorpej #endif
    620  1.28   thorpej 		bufsize = so->so_snd.sb_hiwat;
    621  1.28   thorpej 	if (bufsize < mss)
    622  1.28   thorpej 		mss = bufsize;
    623  1.28   thorpej 	else {
    624  1.28   thorpej 		bufsize = roundup(bufsize, mss);
    625  1.28   thorpej 		if (bufsize > sb_max)
    626  1.28   thorpej 			bufsize = sb_max;
    627  1.28   thorpej 		(void) sbreserve(&so->so_snd, bufsize);
    628  1.28   thorpej 	}
    629  1.33       kml 	tp->t_peermss = mss;
    630  1.33       kml 	tp->t_segsz = mss;
    631  1.28   thorpej 
    632  1.28   thorpej 	/* Initialize the initial congestion window. */
    633  1.36   thorpej 	tp->snd_cwnd = TCP_INITIAL_WINDOW(mss);
    634  1.28   thorpej 
    635  1.28   thorpej #ifdef RTV_SSTHRESH
    636  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    637  1.28   thorpej 		/*
    638  1.28   thorpej 		 * There's some sort of gateway or interface buffer
    639  1.28   thorpej 		 * limit on the path.  Use this to set the slow
    640  1.28   thorpej 		 * start threshold, but set the threshold to no less
    641  1.28   thorpej 		 * than 2 * MSS.
    642  1.28   thorpej 		 */
    643  1.28   thorpej 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    644  1.28   thorpej 	}
    645  1.28   thorpej #endif
    646  1.28   thorpej }
    647  1.28   thorpej 
    648  1.28   thorpej /*
    649  1.28   thorpej  * Processing necessary when a TCP connection is established.
    650  1.28   thorpej  */
    651  1.28   thorpej void
    652  1.28   thorpej tcp_established(tp)
    653  1.28   thorpej 	struct tcpcb *tp;
    654  1.28   thorpej {
    655  1.28   thorpej 	struct inpcb *inp = tp->t_inpcb;
    656  1.28   thorpej 	struct socket *so = inp->inp_socket;
    657  1.28   thorpej #ifdef RTV_RPIPE
    658  1.28   thorpej 	struct rtentry *rt = in_pcbrtentry(inp);
    659  1.28   thorpej #endif
    660  1.28   thorpej 	u_long bufsize;
    661  1.28   thorpej 
    662  1.28   thorpej 	tp->t_state = TCPS_ESTABLISHED;
    663  1.28   thorpej 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    664  1.28   thorpej 
    665  1.28   thorpej #ifdef RTV_RPIPE
    666  1.28   thorpej 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    667  1.28   thorpej 		bufsize = rt->rt_rmx.rmx_recvpipe;
    668  1.28   thorpej 	else
    669  1.28   thorpej #endif
    670  1.28   thorpej 		bufsize = so->so_rcv.sb_hiwat;
    671  1.28   thorpej 	if (bufsize > tp->t_ourmss) {
    672  1.28   thorpej 		bufsize = roundup(bufsize, tp->t_ourmss);
    673  1.28   thorpej 		if (bufsize > sb_max)
    674  1.28   thorpej 			bufsize = sb_max;
    675  1.28   thorpej 		(void) sbreserve(&so->so_rcv, bufsize);
    676  1.28   thorpej 	}
    677  1.28   thorpej }
    678  1.28   thorpej 
    679  1.28   thorpej /*
    680  1.28   thorpej  * Check if there's an initial rtt or rttvar.  Convert from the
    681  1.28   thorpej  * route-table units to scaled multiples of the slow timeout timer.
    682  1.28   thorpej  * Called only during the 3-way handshake.
    683  1.28   thorpej  */
    684  1.28   thorpej void
    685  1.28   thorpej tcp_rmx_rtt(tp)
    686  1.28   thorpej 	struct tcpcb *tp;
    687  1.28   thorpej {
    688  1.28   thorpej #ifdef RTV_RTT
    689  1.28   thorpej 	struct rtentry *rt;
    690  1.28   thorpej 	int rtt;
    691  1.28   thorpej 
    692  1.28   thorpej 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    693  1.28   thorpej 		return;
    694  1.28   thorpej 
    695  1.28   thorpej 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    696  1.28   thorpej 		/*
    697  1.28   thorpej 		 * XXX The lock bit for MTU indicates that the value
    698  1.28   thorpej 		 * is also a minimum value; this is subject to time.
    699  1.28   thorpej 		 */
    700  1.28   thorpej 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    701  1.28   thorpej 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
    702  1.28   thorpej 		tp->t_srtt = rtt /
    703  1.28   thorpej 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    704  1.28   thorpej 		if (rt->rt_rmx.rmx_rttvar) {
    705  1.28   thorpej 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    706  1.28   thorpej 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    707  1.28   thorpej 				(TCP_RTTVAR_SHIFT + 2));
    708  1.28   thorpej 		} else {
    709  1.28   thorpej 			/* Default variation is +- 1 rtt */
    710  1.28   thorpej 			tp->t_rttvar =
    711  1.28   thorpej 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    712  1.28   thorpej 		}
    713  1.28   thorpej 		TCPT_RANGESET(tp->t_rxtcur,
    714  1.28   thorpej 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    715  1.28   thorpej 		    tp->t_rttmin, TCPTV_REXMTMAX);
    716  1.28   thorpej 	}
    717  1.28   thorpej #endif
    718  1.29  explorer }
    719  1.29  explorer 
    720  1.30  explorer tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
    721  1.30  explorer 
    722  1.29  explorer /*
    723  1.29  explorer  * Get a new sequence value given a tcp control block
    724  1.29  explorer  */
    725  1.29  explorer tcp_seq
    726  1.29  explorer tcp_new_iss(tp, len, addin)
    727  1.29  explorer 	void            *tp;
    728  1.29  explorer 	u_long           len;
    729  1.29  explorer 	tcp_seq		 addin;
    730  1.29  explorer {
    731  1.29  explorer 	tcp_seq          tcp_iss;
    732  1.29  explorer 
    733  1.29  explorer 	/*
    734  1.29  explorer 	 * add randomness about this connection, but do not estimate
    735  1.29  explorer 	 * entropy from the timing, since the physical device driver would
    736  1.29  explorer 	 * have done that for us.
    737  1.29  explorer 	 */
    738  1.30  explorer #if NRND > 0
    739  1.29  explorer 	if (tp != NULL)
    740  1.29  explorer 		rnd_add_data(NULL, tp, len, 0);
    741  1.30  explorer #endif
    742  1.29  explorer 
    743  1.29  explorer 	/*
    744  1.29  explorer 	 * randomize.
    745  1.29  explorer 	 */
    746  1.30  explorer #if NRND > 0
    747  1.29  explorer 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    748  1.30  explorer #else
    749  1.30  explorer 	tcp_iss = random();
    750  1.30  explorer #endif
    751  1.29  explorer 
    752  1.29  explorer 	/*
    753  1.29  explorer 	 * If we were asked to add some amount to a known value,
    754  1.29  explorer 	 * we will take a random value obtained above, mask off the upper
    755  1.29  explorer 	 * bits, and add in the known value.  We also add in a constant to
    756  1.29  explorer 	 * ensure that we are at least a certain distance from the original
    757  1.29  explorer 	 * value.
    758  1.29  explorer 	 *
    759  1.29  explorer 	 * This is used when an old connection is in timed wait
    760  1.29  explorer 	 * and we have a new one coming in, for instance.
    761  1.29  explorer 	 */
    762  1.29  explorer 	if (addin != 0) {
    763  1.29  explorer #ifdef TCPISS_DEBUG
    764  1.29  explorer 		printf("Random %08x, ", tcp_iss);
    765  1.29  explorer #endif
    766  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    767  1.30  explorer 		tcp_iss = tcp_iss + addin + TCP_ISSINCR;
    768  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    769  1.30  explorer 		tcp_iss += tcp_iss_seq;
    770  1.29  explorer #ifdef TCPISS_DEBUG
    771  1.29  explorer 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    772  1.29  explorer #endif
    773  1.29  explorer 	} else {
    774  1.29  explorer 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    775  1.30  explorer 		tcp_iss_seq += TCP_ISSINCR;
    776  1.29  explorer 		tcp_iss += tcp_iss_seq;
    777  1.29  explorer #ifdef TCPISS_DEBUG
    778  1.29  explorer 		printf("ISS %08x\n", tcp_iss);
    779  1.29  explorer #endif
    780  1.29  explorer 	}
    781  1.29  explorer 
    782  1.29  explorer #ifdef TCP_COMPAT_42
    783  1.29  explorer 	/*
    784  1.29  explorer 	 * limit it to the positive range for really old TCP implementations
    785  1.29  explorer 	 */
    786  1.29  explorer 	if ((int)tcp_iss < 0)
    787  1.29  explorer 		tcp_iss &= 0x7fffffff;		/* XXX */
    788  1.29  explorer #endif
    789  1.29  explorer 
    790  1.29  explorer 	return tcp_iss;
    791   1.1       cgd }
    792