tcp_output.c

ReactOS是一些高手根据Windows XP的内核编写出的类XP。内核实现机理和API函数调用几乎相同。甚至可以兼容XP的程序。喜欢研究系统内核的人可以看一看。

		/*
		 * This is our initial SYN, check whether we have to use
		 * CC or CC.new.
		 */
		case TH_SYN:
#if 0
			opt[optlen++] = TCPOPT_NOP;
			opt[optlen++] = TCPOPT_NOP;
			opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
						TCPOPT_CCNEW : TCPOPT_CC;
			opt[optlen++] = TCPOLEN_CC;
			*(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
 			optlen += 4;
#endif
			break;

		/*
		 * This is a SYN,ACK; send CC and CC.echo if we received
		 * CC from our peer.
		 */
		case (TH_SYN|TH_ACK):
			if (tp->t_flags & TF_RCVD_CC) {
				opt[optlen++] = TCPOPT_NOP;
				opt[optlen++] = TCPOPT_NOP;
				opt[optlen++] = TCPOPT_CC;
				opt[optlen++] = TCPOLEN_CC;
				*(u_int32_t *)&opt[optlen] =
					htonl(tp->cc_send);
				optlen += 4;
				opt[optlen++] = TCPOPT_NOP;
				opt[optlen++] = TCPOPT_NOP;
				opt[optlen++] = TCPOPT_CCECHO;
				opt[optlen++] = TCPOLEN_CC;
				*(u_int32_t *)&opt[optlen] =
					htonl(tp->cc_recv);
				optlen += 4;
			}
			break;
		}
 	}

 	hdrlen += optlen;

	/*
	 * Adjust data length if insertion of options will
	 * bump the packet length beyond the t_maxopd length.
	 * Clear the FIN bit because we cut off the tail of
	 * the segment.
	 */
	 if (len + optlen > tp->t_maxopd) {
		/*
		 * If there is still more to send, don't close the connection.
		 */
		flags &= ~TH_FIN;
		len = tp->t_maxopd - optlen;
		sendalot = 1;
	}

/*#ifdef DIAGNOSTIC*/
 	if (max_linkhdr + hdrlen > MHLEN)
		panic("tcphdr too big");
/*#endif*/

	/*
	 * Grab a header mbuf, attaching a copy of data to
	 * be transmitted, and initialize the header from
	 * the template for sends on this connection.
	 */
	if (len) {
		if (tp->t_force && len == 1)
			tcpstat.tcps_sndprobe++;
		else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
			tcpstat.tcps_sndrexmitpack++;
			tcpstat.tcps_sndrexmitbyte += len;
		} else {
			tcpstat.tcps_sndpack++;
			tcpstat.tcps_sndbyte += len;
		}
#ifdef notyet
		if ((m = m_copypack(so->so_snd.sb_mb, off,
		    (int)len, max_linkhdr + hdrlen)) == 0) {
			error = ENOBUFS;
			goto out;
		}
		/*
		 * m_copypack left space for our hdr; use it.
		 */
		m->m_len += hdrlen;
		m->m_data -= hdrlen;
#else
		MGETHDR(m, M_DONTWAIT, MT_HEADER);
		if (m == NULL) {
			error = ENOBUFS;
			goto out;
		}
		m->m_data += max_linkhdr;
		m->m_len = hdrlen;
		/* m is not initialized here ... see below up to line
		 * in_cksum to see how it gets there */
		if (len <= MHLEN - hdrlen - max_linkhdr) {
		    m_copydata(so->so_snd.sb_mb, off, (int) len,
			       mtod(m, caddr_t) + hdrlen);
		    m->m_len += len;
		} else {
		    m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
		    // the buffer is allocated, but not filled with the tcp
		    // header yet, so dumping it here yields garbage...
		    //OskitDumpBuffer(mtod(m, caddr_t), len);
		    if (m->m_next == 0) {
			(void) m_free(m);
			error = ENOBUFS;
			goto out;
		    }
		}
#endif
		/*
		 * If we're sending everything we've got, set PUSH.
		 * (This will keep happy those implementations which only
		 * give data to the user when a buffer fills or
		 * a PUSH comes in.)
		 */
		if (off + len == so->so_snd.sb_cc)
			flags |= TH_PUSH;
	} else {
		if (tp->t_flags & TF_ACKNOW)
			tcpstat.tcps_sndacks++;
		else if (flags & (TH_SYN|TH_FIN|TH_RST))
			tcpstat.tcps_sndctrl++;
		else if (SEQ_GT(tp->snd_up, tp->snd_una))
			tcpstat.tcps_sndurg++;
		else
			tcpstat.tcps_sndwinup++;

		MGETHDR(m, M_DONTWAIT, MT_HEADER);
		if (m == NULL) {
			error = ENOBUFS;
			goto out;
		}
		m->m_data += max_linkhdr;
		m->m_len = hdrlen;
	}
	m->m_pkthdr.rcvif = (struct ifnet *)0;

	/* This pulls the data ptr from m and start initting it...
	 * before this point, m is empty. */
	ti = mtod(m, struct tcpiphdr *);
	if (tp->t_template == 0)
		panic("tcp_output");
	(void)memcpy(ti, tp->t_template, sizeof (struct tcpiphdr));

	/*
	 * Fill in fields, remembering maximum advertised
	 * window for use in delaying messages about window sizes.
	 * If resending a FIN, be sure not to use a new sequence number.
	 */
	if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
	    tp->snd_nxt == tp->snd_max)
		tp->snd_nxt--;
	/*
	 * If we are doing retransmissions, then snd_nxt will
	 * not reflect the first unsent octet.  For ACK only
	 * packets, we do not want the sequence number of the
	 * retransmitted packet, we want the sequence number
	 * of the next unsent octet.  So, if there is no data
	 * (and no SYN or FIN), use snd_max instead of snd_nxt
	 * when filling in ti_seq.  But if we are in persist
	 * state, snd_max might reflect one byte beyond the
	 * right edge of the window, so use snd_nxt in that
	 * case, since we know we aren't doing a retransmission.
	 * (retransmit and persist are mutually exclusive...)
	 */
	if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
		ti->ti_seq = htonl(tp->snd_nxt);
	else
		ti->ti_seq = htonl(tp->snd_max);
	ti->ti_ack = htonl(tp->rcv_nxt);

	if (optlen) {
		(void)memcpy(ti + 1, opt, optlen);
		ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
	}
	ti->ti_flags = flags;
	/*
	 * Calculate receive window.  Don't shrink window,
	 * but avoid silly window syndrome.
	 */
	if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
		win = 0;
	if (win > (long)TCP_MAXWIN << tp->rcv_scale)
		win = (long)TCP_MAXWIN << tp->rcv_scale;
	if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
		win = (long)(tp->rcv_adv - tp->rcv_nxt);
	ti->ti_win = htons((u_short) (win>>tp->rcv_scale));
	if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
		ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
		ti->ti_flags |= TH_URG;
	} else
		/*
		 * If no urgent pointer to send, then we pull
		 * the urgent pointer to the left edge of the send window
		 * so that it doesn't drift into the send window on sequence
		 * number wraparound.
		 */
		tp->snd_up = tp->snd_una;		/* drag it along */

	/*
	 * Put TCP length in extended header, and then
	 * checksum extended header and data.
	 */
	if (len + optlen)
		ti->ti_len = htons((u_short)(sizeof (struct tcphdr) +
		    optlen + len));
	ti->ti_sum = in_cksum(m, (int)(hdrlen + len));

	/*
	 * In transmit state, time the transmission and arrange for
	 * the retransmit.  In persist state, just set snd_max.
	 */
	if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
		tcp_seq startseq = tp->snd_nxt;

		/*
		 * Advance snd_nxt over sequence space of this segment.
		 */
		if (flags & (TH_SYN|TH_FIN)) {
			if (flags & TH_SYN)
				tp->snd_nxt++;
			if (flags & TH_FIN) {
				tp->snd_nxt++;
				tp->t_flags |= TF_SENTFIN;
			}
		}
		tp->snd_nxt += len;
		if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
			tp->snd_max = tp->snd_nxt;
			/*
			 * Time this transmission if not a retransmission and
			 * not currently timing anything.
			 */
			if (tp->t_rtt == 0) {
				tp->t_rtt = 1;
				tp->t_rtseq = startseq;
				tcpstat.tcps_segstimed++;
			}
		}

		/*
		 * Set retransmit timer if not currently set,
		 * and not doing an ack or a keep-alive probe.
		 * Initial value for retransmit timer is smoothed
		 * round-trip time + 2 * round-trip time variance.
		 * Initialize shift counter which is used for backoff
		 * of retransmit time.
		 */
		if (tp->t_timer[TCPT_REXMT] == 0 &&
		    tp->snd_nxt != tp->snd_una) {
			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
			if (tp->t_timer[TCPT_PERSIST]) {
				tp->t_timer[TCPT_PERSIST] = 0;
				tp->t_rxtshift = 0;
			}
		}
	} else
		if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
			tp->snd_max = tp->snd_nxt + len;

#ifdef TCPDEBUG
	/*
	 * Trace.
	 */
	if (so->so_options & SO_DEBUG)
		tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0);
#endif

	/*
	 * Fill in IP length and desired time to live and
	 * send to IP level.  There should be a better way
	 * to handle ttl and tos; we could keep them in
	 * the template, but need a way to checksum without them.
	 */
	m->m_pkthdr.len = hdrlen + len;
#ifdef TUBA
	if (tp->t_tuba_pcb)
		error = tuba_output(m, tp);
	else
#endif
    {
#if 1
	struct rtentry *rt;
#endif
	((struct ip *)ti)->ip_len = m->m_pkthdr.len;
	((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl;	/* XXX */
	((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos;	/* XXX */
#if 1
	/*
	 * See if we should do MTU discovery.  We do it only if the following
	 * are true:
	 *	1) we have a valid route to the destination
	 *	2) the MTU is not locked (if it is, then discovery has been
	 *	   disabled)
	 */
	if ((rt = tp->t_inpcb->inp_route.ro_rt)
	    && rt->rt_flags & RTF_UP
	    && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
		((struct ip *)ti)->ip_off |= IP_DF;
	}
#endif
        /*
         * XXX: It seems that osktittcp expects that packets are
         * synchronously processed. The current implementation feeds
         * oskittcp with the packets asynchronously. That's not a
         * problem normally when the packets are transfered over
         * network, but it starts to be a problem when it comes to
         * loopback packets.
         * The ACK bits are set in tcp_input which calls tcp_output and
         * expects them to be cleared before further processing.
         * Instead tcp_output calls ip_output which produces a packet
         * and ends up in tcp_input and we're stuck in infinite loop.
         * Normally the flags are masked out at the end of this function
         * and the incomming packets are processed then, but since
         * currently the loopback packet is delivered during the
         * ip_output call, the function end is never reached...
         */
#ifdef __REACTOS__
	tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
#endif
	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
			  so->so_options & SO_DONTROUTE, 0);
#ifdef __REACTOS__
	/* We allocated m, so we are responsible for freeing it. If the mbuf
           contains a pointer to an external datablock, we (or rather, m_copy)
           didn't allocate it but pointed it to the data to send. So we have
           to cheat a little bit and keep M_FREE from freeing the external
           data block */
	while (NULL != m) {
	    m->m_flags &= ~M_EXT;
	    MFREE(m, n);
	    m = n;
	}
#endif
    }
	if (error) {
out:
		if (error == ENOBUFS) {
			tcp_quench(tp->t_inpcb, 0);
			OS_DbgPrint(OSK_MID_TRACE,("quench 0\n"));
			return (0);
		}
#if 1
		if (error == EMSGSIZE) {
			/*
			 * ip_output() will have already fixed the route
			 * for us.  tcp_mtudisc() will, as its last action,
			 * initiate retransmission, so it is important to
			 * not do so here.
			 */
			tcp_mtudisc(tp->t_inpcb, 0);
			OS_DbgPrint(OSK_MID_TRACE,("mtudisc 0\n"));
			return 0;
		}
#endif
		if ((error == EHOSTUNREACH || error == ENETDOWN)
		    && TCPS_HAVERCVDSYN(tp->t_state)) {
			tp->t_softerror = error;
			OS_DbgPrint(OSK_MID_TRACE,("softerror %d\n", error));
			return (0);
		}
		OS_DbgPrint(OSK_MID_TRACE,("error %d\n", error));
		return (error);
	}
	tcpstat.tcps_sndtotal++;

	/*
	 * Data sent (as far as we can tell).
	 * If this advertises a larger window than any other segment,
	 * then remember the size of the advertised window.
	 * Any pending ACK has now been sent.
	 */
	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
		tp->rcv_adv = tp->rcv_nxt + win;
	tp->last_ack_sent = tp->rcv_nxt;
	tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
	if (sendalot)
		goto again;
	OS_DbgPrint(OSK_MID_TRACE,("leaving 0\n"));
	return (0);
}

void
tcp_setpersist(tp)
	register struct tcpcb *tp;
{
	register int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;

	if (tp->t_timer[TCPT_REXMT])
		panic("tcp_output REXMT");
	/*
	 * Start/restart persistance timer.
	 */
	TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
	    t * tcp_backoff[tp->t_rxtshift],
	    TCPTV_PERSMIN, TCPTV_PERSMAX);
	if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
		tp->t_rxtshift++;
}