tcp_send.c

minix3的源代码

/*
tcp_send.c

Copyright 1995 Philip Homburg
*/

#include "inet.h"
#include "buf.h"
#include "clock.h"
#include "event.h"
#include "type.h"
#include "sr.h"

#include "assert.h"
#include "io.h"
#include "ip.h"
#include "tcp.h"
#include "tcp_int.h"

THIS_FILE

FORWARD acc_t *make_pack ARGS(( tcp_conn_t *tcp_conn ));
FORWARD void tcp_send_timeout ARGS(( int conn, struct timer *timer ));
FORWARD void do_snd_event ARGS(( event_t *ev, ev_arg_t arg ));

PUBLIC void tcp_conn_write (tcp_conn, enq)
tcp_conn_t *tcp_conn;
int enq;				/* Writes need to be enqueued. */
{
	tcp_port_t *tcp_port;
	ev_arg_t snd_arg;

	assert (tcp_conn->tc_flags & TCF_INUSE);

	tcp_port= tcp_conn->tc_port;
	if (tcp_conn->tc_flags & TCF_MORE2WRITE)
		return;

	/* Do we really have something to send here? */
	if (tcp_conn->tc_SND_UNA == tcp_conn->tc_SND_NXT &&
		!(tcp_conn->tc_flags & TCF_SEND_ACK) &&
		!tcp_conn->tc_frag2send)
	{
		return;
	}

	tcp_conn->tc_flags |= TCF_MORE2WRITE;
	tcp_conn->tc_send_link= NULL;
	if (!tcp_port->tp_snd_head)
	{
		tcp_port->tp_snd_head= tcp_conn;
		tcp_port->tp_snd_tail= tcp_conn;
		if (enq)
		{
			snd_arg.ev_ptr= tcp_port;
			if (!ev_in_queue(&tcp_port->tp_snd_event))
			{
				ev_enqueue(&tcp_port->tp_snd_event,
					do_snd_event, snd_arg);
			}
		}
		else
			tcp_port_write(tcp_port);
	}
	else
	{
		tcp_port->tp_snd_tail->tc_send_link= tcp_conn;
		tcp_port->tp_snd_tail= tcp_conn;
	}
}

PRIVATE void do_snd_event(ev, arg)
event_t *ev;
ev_arg_t arg;
{
	tcp_port_t *tcp_port;

	tcp_port= arg.ev_ptr;

	assert(ev == &tcp_port->tp_snd_event);
	tcp_port_write(tcp_port);
}

PUBLIC void tcp_port_write(tcp_port)
tcp_port_t *tcp_port;
{
	tcp_conn_t *tcp_conn;
	acc_t *pack2write;
	int r;

	assert (!(tcp_port->tp_flags & TPF_WRITE_IP));

	while(tcp_port->tp_snd_head)
	{
		tcp_conn= tcp_port->tp_snd_head;
		assert(tcp_conn->tc_flags & TCF_MORE2WRITE);

		for(;;)
		{
			if (tcp_conn->tc_frag2send)
			{
				pack2write= tcp_conn->tc_frag2send;
				tcp_conn->tc_frag2send= 0;
			}
			else
			{
				tcp_conn->tc_busy++;
				pack2write= make_pack(tcp_conn);
				tcp_conn->tc_busy--;
				if (!pack2write)
					break;
			}
			r= ip_send(tcp_port->tp_ipfd, pack2write,
				bf_bufsize(pack2write));
			if (r != NW_OK)
			{
				if (r == NW_WOULDBLOCK)
					break;
				if (r == EPACKSIZE)
				{
					tcp_mtu_exceeded(tcp_conn);
					continue;
				}
				if (r == EDSTNOTRCH)
				{
					tcp_notreach(tcp_conn);
					continue;
				}
				if (r == EBADDEST)
					continue;
			}
			assert(r == NW_OK ||
				(printf("ip_send failed, error %d\n", r),0));
		}

		if (pack2write)
		{
			tcp_port->tp_flags |= TPF_WRITE_IP;
			tcp_port->tp_pack= pack2write;

			r= ip_write (tcp_port->tp_ipfd,
				bf_bufsize(pack2write));
			if (r == NW_SUSPEND)
			{
				tcp_port->tp_flags |= TPF_WRITE_SP;
				return;
			}
			assert(r == NW_OK);
			tcp_port->tp_flags &= ~TPF_WRITE_IP;
			assert(!(tcp_port->tp_flags &
				(TPF_WRITE_IP|TPF_WRITE_SP)));
			continue;
		}
		tcp_conn->tc_flags &= ~TCF_MORE2WRITE;
		tcp_port->tp_snd_head= tcp_conn->tc_send_link;

	}
}

PRIVATE acc_t *make_pack(tcp_conn)
tcp_conn_t *tcp_conn;
{
	acc_t *pack2write, *tmp_pack, *tcp_pack;
	tcp_hdr_t *tcp_hdr;
	ip_hdr_t *ip_hdr;
	int tot_hdr_size, ip_hdr_len, no_push, head, more2write;
	u32_t seg_seq, seg_lo_data, queue_lo_data, seg_hi, seg_hi_data;
	u16_t seg_up, mss;
	u8_t seg_flags;
	size_t pack_size;
	clock_t curr_time, new_dis;
	u8_t *optptr;

	mss= tcp_conn->tc_mtu-IP_TCP_MIN_HDR_SIZE;

	assert(tcp_conn->tc_busy);
	curr_time= get_time();
	switch (tcp_conn->tc_state)
	{
	case TCS_CLOSED:
	case TCS_LISTEN:
		return NULL;
	case TCS_SYN_RECEIVED:
	case TCS_SYN_SENT:

		if (tcp_conn->tc_SND_TRM == tcp_conn->tc_SND_NXT &&
			!(tcp_conn->tc_flags & TCF_SEND_ACK))
		{
			return 0;
		}

		tcp_conn->tc_flags &= ~TCF_SEND_ACK;

		/* Advertise a mss based on the port mtu. The current mtu may
		 * be lower if the other side sends a smaller mss.
		 */
		mss= tcp_conn->tc_port->tp_mtu-IP_TCP_MIN_HDR_SIZE;

		/* Include a max segment size option. */
		assert(tcp_conn->tc_tcpopt == NULL);
		tcp_conn->tc_tcpopt= bf_memreq(4);
		optptr= (u8_t *)ptr2acc_data(tcp_conn->tc_tcpopt);
		optptr[0]= TCP_OPT_MSS;
		optptr[1]= 4;
		optptr[2]= mss >> 8;
		optptr[3]= mss & 0xFF;

		pack2write= tcp_make_header(tcp_conn, &ip_hdr, &tcp_hdr, 
			(acc_t *)0);

		bf_afree(tcp_conn->tc_tcpopt);
		tcp_conn->tc_tcpopt= NULL;

		if (!pack2write)
		{
			DBLOCK(1, printf("connection closed while inuse\n"));
			return 0;
		}
		tot_hdr_size= bf_bufsize(pack2write);
		seg_seq= tcp_conn->tc_SND_TRM;
		if (tcp_conn->tc_state == TCS_SYN_SENT)
			seg_flags= 0;
		else
			seg_flags= THF_ACK;	/* except for TCS_SYN_SENT
						 * ack is always present */

		if (seg_seq == tcp_conn->tc_ISS)
		{
			assert(tcp_conn->tc_transmit_timer.tim_active ||
				(tcp_print_conn(tcp_conn), printf("\n"), 0));
			seg_flags |= THF_SYN;
			tcp_conn->tc_SND_TRM++;
		}

		tcp_hdr->th_seq_nr= htonl(seg_seq);
		tcp_hdr->th_ack_nr= htonl(tcp_conn->tc_RCV_NXT);
		tcp_hdr->th_flags= seg_flags;
		tcp_hdr->th_window= htons(mss);
			/* Initially we allow one segment */

		ip_hdr->ih_length= htons(tot_hdr_size);

		pack2write->acc_linkC++;
		ip_hdr_len= (ip_hdr->ih_vers_ihl & IH_IHL_MASK) << 2;
		tcp_pack= bf_delhead(pack2write, ip_hdr_len);
		tcp_hdr->th_chksum= ~tcp_pack_oneCsum(ip_hdr, tcp_pack);
		bf_afree(tcp_pack);

		new_dis= curr_time + 2*HZ*tcp_conn->tc_ttl;
		if (new_dis > tcp_conn->tc_senddis)
			tcp_conn->tc_senddis= new_dis;
		return pack2write;

	case TCS_ESTABLISHED:
	case TCS_CLOSING:
		seg_seq= tcp_conn->tc_SND_TRM;

		seg_flags= 0;
		pack2write= 0;
		seg_up= 0;
		if (tcp_conn->tc_flags & TCF_SEND_ACK)
		{
			seg_flags= THF_ACK;
			tcp_conn->tc_flags &= ~TCF_SEND_ACK;

			pack2write= tcp_make_header (tcp_conn, &ip_hdr, 
				&tcp_hdr, (acc_t *)0);
			if (!pack2write)
			{
				return NULL;
			}
		}

		if (tcp_conn->tc_SND_UNA != tcp_conn->tc_SND_NXT)
		{
			assert(tcp_LEmod4G(seg_seq, tcp_conn->tc_SND_NXT));

			if (seg_seq == tcp_conn->tc_snd_cwnd)
			{
				DBLOCK(2,
					printf("no data: window is closed\n"));
				goto after_data;
			}

			/* Assert that our SYN has been ACKed. */
			assert(tcp_conn->tc_SND_UNA != tcp_conn->tc_ISS);

			seg_lo_data= seg_seq;
			queue_lo_data= tcp_conn->tc_SND_UNA;

			seg_hi= tcp_conn->tc_SND_NXT;
			seg_hi_data= seg_hi;
			if (tcp_conn->tc_flags & TCF_FIN_SENT)
			{
				if (seg_seq != seg_hi)
					seg_flags |= THF_FIN;
				if (queue_lo_data == seg_hi_data)
					queue_lo_data--;
				if (seg_lo_data == seg_hi_data)
					seg_lo_data--;
				seg_hi_data--;
			}

			if (!pack2write)
			{
				pack2write= tcp_make_header (tcp_conn,
					&ip_hdr, &tcp_hdr, (acc_t *)0);
				if (!pack2write)
				{
					return NULL;
				}
			}

			tot_hdr_size= bf_bufsize(pack2write);

			no_push= (tcp_LEmod4G(tcp_conn->tc_SND_PSH, seg_seq));
			head= (seg_seq == tcp_conn->tc_SND_UNA);
			if (no_push)
			{
				/* Shutdown sets SND_PSH */
				seg_flags &= ~THF_FIN;
				if (seg_hi_data-seg_lo_data <= 1)
				{
					/* Allways keep at least one byte
					 * for a future push.
					 */
					DBLOCK(0x20,
					    printf("no data: no push\n"));
					if (head)
					{
						DBLOCK(0x1, printf(
					"no data: setting TCF_NO_PUSH\n"));
						tcp_conn->tc_flags |=
							TCF_NO_PUSH;
					}
					goto after_data;
				}
				seg_hi_data--;
			}

			if (tot_hdr_size != IP_TCP_MIN_HDR_SIZE)
			{
				printf(
				"tcp_write`make_pack: tot_hdr_size = %d\n",
					tot_hdr_size);
				mss= tcp_conn->tc_mtu-tot_hdr_size;
			}
			if (seg_hi_data - seg_lo_data > mss)
			{
				/* Truncate to at most one segment */
				seg_hi_data= seg_lo_data + mss;
				seg_hi= seg_hi_data;
				seg_flags &= ~THF_FIN;
			}

			if (no_push &&
				seg_hi_data-seg_lo_data != mss)
			{
				DBLOCK(0x20, printf(
				"no data: no push for partial segment\n"));
				more2write= (tcp_conn->tc_fd &&
					(tcp_conn->tc_fd->tf_flags &
					TFF_WRITE_IP));
				DIFBLOCK(2, more2write, 
					printf(
			"tcp_send`make_pack: more2write -> !TCF_NO_PUSH\n");
				);
				if (head && !more2write)
				{
					DBLOCK(0x1, printf(
				"partial segment: setting TCF_NO_PUSH\n"));
					tcp_conn->tc_flags |= TCF_NO_PUSH;
					tcp_print_conn(tcp_conn);
					printf("\n");
				}
				goto after_data;
			}


			if (tcp_Gmod4G(seg_hi, tcp_conn->tc_snd_cwnd))
			{
				seg_hi_data= tcp_conn->tc_snd_cwnd;
				seg_hi= seg_hi_data;
				seg_flags &= ~THF_FIN;
			}

			if (!head &&
				seg_hi_data-seg_lo_data < mss)
			{
				if (tcp_conn->tc_flags & TCF_PUSH_NOW)
				{
					DBLOCK(0x20,
					printf("push: no Nagle\n"));
				}
				else
				{
				DBLOCK(0x20,
					printf("no data: partial packet\n"));
				seg_flags &= ~THF_FIN;
				goto after_data;
				}
			}

			if (seg_hi-seg_seq == 0)
			{
				DBLOCK(0x20,
				printf("no data: no data available\n"));
				goto after_data;
			}

			if (tcp_GEmod4G(tcp_conn->tc_SND_UP, seg_lo_data))
			{
				extern int killer_inet;

				if (tcp_GEmod4G(tcp_conn->tc_SND_UP,
					seg_hi_data))
				{
					seg_up= seg_hi_data-seg_seq;
				}
				else
				{
					seg_up= tcp_conn->tc_SND_UP-seg_seq;
				}
				seg_flags |= THF_URG;
				if (!killer_inet &&
					(tcp_conn->tc_flags & TCF_BSD_URG) &&
					seg_up == 0)
				{
					/* A zero urgent pointer doesn't mean
					 * anything when BSD semantics are
					 * used (urgent pointer points to the
					 * first no urgent byte). The use of
					 * a zero urgent pointer also crashes
					 * a Solaris 2.3 kernel. If urgent
					 * pointer doesn't have BSD semantics
					 * then an urgent pointer of zero
					 * simply indicates that there is one
					 * urgent byte.
					 */
					seg_flags &= ~THF_URG;
				}
			}
			else
				seg_up= 0;

			if (tcp_Gmod4G(tcp_conn->tc_SND_PSH, seg_lo_data) &&
				tcp_LEmod4G(tcp_conn->tc_SND_PSH, seg_hi_data))
			{
				seg_flags |= THF_PSH;
			}

			tcp_conn->tc_SND_TRM= seg_hi;

			assert(tcp_conn->tc_transmit_timer.tim_active ||
				(tcp_print_conn(tcp_conn), printf("\n"), 0));
			if (tcp_conn->tc_rt_seq == 0 && 
				tcp_Gmod4G(seg_seq, tcp_conn->tc_rt_threshold))
			{
				tcp_conn->tc_rt_time= curr_time;
				tcp_conn->tc_rt_seq=
					tcp_conn->tc_rt_threshold= seg_seq;
			}

			if (seg_hi_data-seg_lo_data)
			{
#if DEBUG & 0
				assert(tcp_check_conn(tcp_conn));
				assert((seg_hi_data-queue_lo_data <=
					bf_bufsize(tcp_conn->tc_send_data) &&
					seg_lo_data-queue_lo_data <=
					bf_bufsize(tcp_conn->tc_send_data) &&
					seg_hi_data>seg_lo_data)||
					(tcp_print_conn(tcp_conn),
					printf(
		" seg_hi_data= 0x%x, seg_lo_data= 0x%x, queue_lo_data= 0x%x\n",
					seg_hi_data, seg_lo_data,
					queue_lo_data), 0));
#endif

				tmp_pack= pack2write;
				while (tmp_pack->acc_next)
					tmp_pack= tmp_pack->acc_next;