tcp_subr.c

eCos/RedBoot for勤研ARM AnywhereII(4510) 含全部源代码

	} else
#endif /* INET6 */
      {
        nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
	    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
        m->m_pkthdr.csum_flags = CSUM_TCP;
        m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
      }
#ifdef TCPDEBUG
	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
#endif
#ifdef IPSEC
	if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
		m_freem(m);
		return;
	}
#endif
#ifdef INET6
	if (isipv6) {
		(void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
		if (ro6 == &sro6 && ro6->ro_rt) {
			RTFREE(ro6->ro_rt);
			ro6->ro_rt = NULL;
		}
	} else
#endif /* INET6 */
      {
	(void) ip_output(m, NULL, ro, ipflags, NULL);
	if (ro == &sro && ro->ro_rt) {
		RTFREE(ro->ro_rt);
		ro->ro_rt = NULL;
	}
      }
}

/*
 * Create a new TCP control block, making an
 * empty reassembly queue and hooking it to the argument
 * protocol control block.  The `inp' parameter must have
 * come from the zone allocator set up in tcp_init().
 */
struct tcpcb *
tcp_newtcpcb(inp)
	struct inpcb *inp;
{
	struct inp_tp *it;
	register struct tcpcb *tp;
#ifdef INET6
	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
#endif /* INET6 */

	it = (struct inp_tp *)inp;
	tp = &it->tcb;
	bzero((char *) tp, sizeof(struct tcpcb));
	LIST_INIT(&tp->t_segq);
	tp->t_maxseg = tp->t_maxopd =
#ifdef INET6
		isipv6 ? tcp_v6mssdflt :
#endif /* INET6 */
		tcp_mssdflt;

	/* Set up our timeouts. */
	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
	callout_init(tp->tt_persist = &it->inp_tp_persist);
	callout_init(tp->tt_keep = &it->inp_tp_keep);
	callout_init(tp->tt_2msl = &it->inp_tp_2msl);
	callout_init(tp->tt_delack = &it->inp_tp_delack);

	if (tcp_do_rfc1323)
		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
	if (tcp_do_rfc1644)
		tp->t_flags |= TF_REQ_CC;
	tp->t_inpcb = inp;	/* XXX */
	/*
	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
	 * reasonable initial retransmit time.
	 */
	tp->t_srtt = TCPTV_SRTTBASE;
	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
	tp->t_rttmin = tcp_rexmit_min;
	tp->t_rxtcur = TCPTV_RTOBASE;
	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
	tp->t_rcvtime = ticks;
        /*
	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
	 * because the socket may be bound to an IPv6 wildcard address,
	 * which may match an IPv4-mapped IPv6 address.
	 */
	inp->inp_ip_ttl = ip_defttl;
	inp->inp_ppcb = (caddr_t)tp;
	return (tp);		/* XXX */
}

/*
 * Drop a TCP connection, reporting
 * the specified error.  If connection is synchronized,
 * then send a RST to peer.
 */
struct tcpcb *
tcp_drop(tp, _errno)
	register struct tcpcb *tp;
	int _errno;
{
	struct socket *so = tp->t_inpcb->inp_socket;

	if (TCPS_HAVERCVDSYN(tp->t_state)) {
		tp->t_state = TCPS_CLOSED;
		(void) tcp_output(tp);
		tcpstat.tcps_drops++;
	} else
		tcpstat.tcps_conndrops++;
	if (_errno == ETIMEDOUT && tp->t_softerror)
		_errno = tp->t_softerror;
	so->so_error = _errno;
	return (tcp_close(tp));
}

/*
 * Close a TCP control block:
 *	discard all space held by the tcp
 *	discard internet protocol block
 *	wake up any sleepers
 */
struct tcpcb *
tcp_close(tp)
	register struct tcpcb *tp;
{
	register struct tseg_qent *q;
	struct inpcb *inp = tp->t_inpcb;
	struct socket *so = inp->inp_socket;
#ifdef INET6
	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
#endif /* INET6 */
	register struct rtentry *rt;
	int dosavessthresh;

	/*
	 * Make sure that all of our timers are stopped before we
	 * delete the PCB.
	 */
	callout_stop(tp->tt_rexmt);
	callout_stop(tp->tt_persist);
	callout_stop(tp->tt_keep);
	callout_stop(tp->tt_2msl);
	callout_stop(tp->tt_delack);

	/*
	 * If we got enough samples through the srtt filter,
	 * save the rtt and rttvar in the routing entry.
	 * 'Enough' is arbitrarily defined as the 16 samples.
	 * 16 samples is enough for the srtt filter to converge
	 * to within 5% of the correct value; fewer samples and
	 * we could save a very bogus rtt.
	 *
	 * Don't update the default route's characteristics and don't
	 * update anything that the user "locked".
	 */
	if (tp->t_rttupdated >= 16) {
		register u_long i = 0;
#ifdef INET6
		if (isipv6) {
			struct sockaddr_in6 *sin6;

			if ((rt = inp->in6p_route.ro_rt) == NULL)
				goto no_valid_rt;
			sin6 = (struct sockaddr_in6 *)rt_key(rt);
			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
				goto no_valid_rt;
		}
		else
#endif /* INET6 */		
		if ((rt = inp->inp_route.ro_rt) == NULL ||
		    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
		    == INADDR_ANY)
			goto no_valid_rt;

		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
			i = tp->t_srtt *
			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
			if (rt->rt_rmx.rmx_rtt && i)
				/*
				 * filter this update to half the old & half
				 * the new values, converting scale.
				 * See route.h and tcp_var.h for a
				 * description of the scaling constants.
				 */
				rt->rt_rmx.rmx_rtt =
				    (rt->rt_rmx.rmx_rtt + i) / 2;
			else
				rt->rt_rmx.rmx_rtt = i;
			tcpstat.tcps_cachedrtt++;
		}
		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
			i = tp->t_rttvar *
			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
			if (rt->rt_rmx.rmx_rttvar && i)
				rt->rt_rmx.rmx_rttvar =
				    (rt->rt_rmx.rmx_rttvar + i) / 2;
			else
				rt->rt_rmx.rmx_rttvar = i;
			tcpstat.tcps_cachedrttvar++;
		}
		/*
		 * The old comment here said:
		 * update the pipelimit (ssthresh) if it has been updated
		 * already or if a pipesize was specified & the threshhold
		 * got below half the pipesize.  I.e., wait for bad news
		 * before we start updating, then update on both good
		 * and bad news.
		 *
		 * But we want to save the ssthresh even if no pipesize is
		 * specified explicitly in the route, because such
		 * connections still have an implicit pipesize specified
		 * by the global tcp_sendspace.  In the absence of a reliable
		 * way to calculate the pipesize, it will have to do.
		 */
		i = tp->snd_ssthresh;
		if (rt->rt_rmx.rmx_sendpipe != 0)
			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
		else
			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
		    || dosavessthresh) {
			/*
			 * convert the limit from user data bytes to
			 * packets then to packet data bytes.
			 */
			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
			if (i < 2)
				i = 2;
			i *= (u_long)(tp->t_maxseg +
#ifdef INET6
				      (isipv6 ? sizeof (struct ip6_hdr) +
					       sizeof (struct tcphdr) :
#endif
				       sizeof (struct tcpiphdr)
#ifdef INET6
				       )
#endif
				      );
			if (rt->rt_rmx.rmx_ssthresh)
				rt->rt_rmx.rmx_ssthresh =
				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
			else
				rt->rt_rmx.rmx_ssthresh = i;
			tcpstat.tcps_cachedssthresh++;
		}
	}
	rt = inp->inp_route.ro_rt;
	if (rt) {
		/* 
		 * mark route for deletion if no information is
		 * cached.
		 */
		if ((tp->t_flags & TF_LQ_OVERFLOW) &&
		    ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){
			if (rt->rt_rmx.rmx_rtt == 0)
				rt->rt_flags |= RTF_DELCLONE;
		}
	}
    no_valid_rt:
	/* free the reassembly queue, if any */
	while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
		LIST_REMOVE(q, tqe_q);
		m_freem(q->tqe_m);
		FREE(q, M_TSEGQ);
	}
	inp->inp_ppcb = NULL;
	soisdisconnected(so);
#ifdef INET6
	if (INP_CHECK_SOCKAF(so, AF_INET6))
		in6_pcbdetach(inp);
	else
#endif /* INET6 */
	in_pcbdetach(inp);
	tcpstat.tcps_closed++;
	return ((struct tcpcb *)0);
}

void
tcp_drain()
{
	if (do_tcpdrain)
	{
		struct inpcb *inpb;
		struct tcpcb *tcpb;
		struct tseg_qent *te;

	/*
	 * Walk the tcpbs, if existing, and flush the reassembly queue,
	 * if there is one...
	 * XXX: The "Net/3" implementation doesn't imply that the TCP
	 *      reassembly queue should be flushed, but in a situation
	 * 	where we're really low on mbufs, this is potentially
	 *  	usefull.	
	 */
		for (inpb = LIST_FIRST(tcbinfo.listhead); inpb;
	    		inpb = LIST_NEXT(inpb, inp_list)) {
				if ((tcpb = intotcpcb(inpb))) {
					while ((te = LIST_FIRST(&tcpb->t_segq))
					       != NULL) {
					LIST_REMOVE(te, tqe_q);
					m_freem(te->tqe_m);
					FREE(te, M_TSEGQ);
				}
			}
		}

	}
}

/*
 * Notify a tcp user of an asynchronous error;
 * store error as soft error, but wake up user
 * (for now, won't do anything until can select for soft error).
 *
 * Do not wake up user since there currently is no mechanism for
 * reporting soft errors (yet - a kqueue filter may be added).
 */
static void
tcp_notify(inp, error)
	struct inpcb *inp;
	int error;
{
	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;

	/*
	 * Ignore some errors if we are hooked up.
	 * If connection hasn't completed, has retransmitted several times,
	 * and receives a second error, give up now.  This is better
	 * than waiting a long time to establish a connection that
	 * can never complete.
	 */
	if (tp->t_state == TCPS_ESTABLISHED &&
	     (error == EHOSTUNREACH || error == ENETUNREACH ||
	      error == EHOSTDOWN)) {
		return;
	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
	    tp->t_softerror)
		tcp_drop(tp, error);
	else
		tp->t_softerror = error;
#if 0
	wakeup((caddr_t) &so->so_timeo);
	sorwakeup(so);
	sowwakeup(so);
#endif
}

#ifdef CYGPKG_NET_FREEBSD_SYSCTL
static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)
{
	int error, i, n, s;
	struct inpcb *inp, **inp_list;
	inp_gen_t gencnt;
	struct xinpgen xig;

	/*
	 * The process of preparing the TCB list is too time-consuming and
	 * resource-intensive to repeat twice on every request.
	 */
	if (req->oldptr == 0) {
		n = tcbinfo.ipi_count;
		req->oldidx = 2 * (sizeof xig)
			+ (n + n/8) * sizeof(struct xtcpcb);
		return 0;
	}

	if (req->newptr != 0)
		return EPERM;

	/*
	 * OK, now we're committed to doing something.
	 */
	s = splnet();
	gencnt = tcbinfo.ipi_gencnt;
	n = tcbinfo.ipi_count;
	splx(s);

	xig.xig_len = sizeof xig;
	xig.xig_count = n;
	xig.xig_gen = gencnt;
	xig.xig_sogen = so_gencnt;
	error = SYSCTL_OUT(req, &xig, sizeof xig);
	if (error)
		return error;

	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
	if (inp_list == 0)
		return ENOMEM;
	
	s = splnet();
	for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
	     inp = LIST_NEXT(inp, inp_list)) {
		if (inp->inp_gencnt <= gencnt)
			inp_list[i++] = inp;
	}
	splx(s);
	n = i;

	error = 0;
	for (i = 0; i < n; i++) {
		inp = inp_list[i];
		if (inp->inp_gencnt <= gencnt) {
			struct xtcpcb xt;
			caddr_t inp_ppcb;
			xt.xt_len = sizeof xt;
			/* XXX should avoid extra copy */
			bcopy(inp, &xt.xt_inp, sizeof *inp);
			inp_ppcb = inp->inp_ppcb;
			if (inp_ppcb != NULL)
				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
			else
				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
			if (inp->inp_socket)
				sotoxsocket(inp->inp_socket, &xt.xt_socket);
			error = SYSCTL_OUT(req, &xt, sizeof xt);
		}
	}
	if (!error) {
		/*
		 * Give the user an updated idea of our state.
		 * If the generation differs from what we told
		 * her before, she knows that something happened
		 * while we were processing this request, and it
		 * might be necessary to retry.
		 */
		s = splnet();
		xig.xig_gen = tcbinfo.ipi_gencnt;
		xig.xig_sogen = so_gencnt;
		xig.xig_count = tcbinfo.ipi_count;
		splx(s);
		error = SYSCTL_OUT(req, &xig, sizeof xig);
	}
	free(inp_list, M_TEMP);
	return error;
}
#endif

SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");

void
tcp_ctlinput(cmd, sa, vip)
	int cmd;
	struct sockaddr *sa;
	void *vip;
{
	struct ip *ip = vip;
	struct tcphdr *th;
	struct in_addr faddr;
	struct inpcb *inp;
	struct tcpcb *tp;
	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
	tcp_seq icmp_seq;
	int s;

	faddr = ((struct sockaddr_in *)sa)->sin_addr;
	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
		return;