tcpin.c

uCLinux下的一个TCP/IP协议栈源码

/* Process incoming TCP segments. Page number references are to ARPA RFC-793,
 * the TCP specification.
 *
 */
#include "global.h"
#include "timer.h"
#include "mbuf.h"
#include "netuser.h"
#include "internet.h"
#include "tcp.h"
#include "icmp.h"
#include "iface.h"
#include "ip.h"

static void update(struct tcb *tcb,struct tcp *seg,uint16 length);
static void proc_syn(struct tcb *tcb,uint8 tos,struct tcp *seg);
static void add_reseq(struct tcb *tcb,uint8 tos,struct tcp *seg,
	struct mbuf **bp,uint16 length);
static void get_reseq(struct tcb *tcb,uint8 *tos,struct tcp *seq,
	struct mbuf **bp,uint16 *length);
static int trim(struct tcb *tcb,struct tcp *seg,struct mbuf **bpp,
	uint16 *length);
static int in_window(struct tcb *tcb,int32 seq);

/* This function is called from IP with the IP header in machine byte order,
 * along with a mbuf chain pointing to the TCP header.
 */
void
tcp_input(
struct iface *iface,	/* Incoming interface (ignored) */
struct ip *ip,		/* IP header */
struct mbuf **bpp,	/* Data field, if any */
int rxbroadcast,	/* Incoming broadcast - discard if true */
int32 said		/* Authenticated packet */
){
	struct tcb *ntcb;
	register struct tcb *tcb;	/* TCP Protocol control block */
	struct tcp seg;			/* Local copy of segment header */
	struct connection conn;		/* Local copy of addresses */
	struct pseudo_header ph;	/* Pseudo-header for checksumming */
	int hdrlen;			/* Length of TCP header */
	uint16 length;
	int32 t;

	if(bpp == NULL || *bpp == NULL)
		return;

	tcpInSegs++;
	if(rxbroadcast){
		/* Any TCP packet arriving as a broadcast is
		 * to be completely IGNORED!!
		 */
		free_p(bpp);
		return;
	}
	length = ip->length - IPLEN - ip->optlen;
	ph.source = ip->source;
	ph.dest = ip->dest;
	ph.protocol = ip->protocol;
	ph.length = length;
	if(cksum(&ph,*bpp,length) != 0){
		/* Checksum failed, ignore segment completely */
		tcpInErrs++;
		free_p(bpp);
		return;
	}
	/* Form local copy of TCP header in host byte order */
	if((hdrlen = ntohtcp(&seg,bpp)) < 0){
		/* TCP header is too small */
		free_p(bpp);
		return;
	}
	length -= hdrlen;

	/* Fill in connection structure and find TCB */
	conn.local.address = ip->dest;
	conn.local.port = seg.dest;
	conn.remote.address = ip->source;
	conn.remote.port = seg.source;
	
	if((tcb = lookup_tcb(&conn)) == NULL){
		/* If this segment doesn't carry a SYN, reject it */
		if(!seg.flags.syn){
			free_p(bpp);
			reset(ip,&seg);
			return;
		}
		/* See if there's a TCP_LISTEN on this socket with
		 * unspecified remote address and port
		 */
		conn.remote.address = 0;
		conn.remote.port = 0;
		if((tcb = lookup_tcb(&conn)) == NULL){
			/* Nope, try unspecified local address too */
			conn.local.address = 0;
			if((tcb = lookup_tcb(&conn)) == NULL){
				/* No LISTENs, so reject */
				free_p(bpp);
				reset(ip,&seg);
				return;
			}
		}
		/* We've found an server listen socket, so clone the TCB */
		if(tcb->flags.clone){
			ntcb = (struct tcb *)mallocw(sizeof (struct tcb));
			ASSIGN(*ntcb,*tcb);
			tcb = ntcb;
			tcb->timer.arg = tcb;
			/* Put on list */
			tcb->next = Tcbs;
			Tcbs = tcb;
		}
		/* Put all the socket info into the TCB */
		tcb->conn.local.address = ip->dest;
		tcb->conn.remote.address = ip->source;
		tcb->conn.remote.port = seg.source;
	}
	tcb->flags.congest = ip->flags.congest;
	/* Do unsynchronized-state processing (p. 65-68) */
	switch(tcb->state){
	case TCP_CLOSED:
		free_p(bpp);
		reset(ip,&seg);
		return;
	case TCP_LISTEN:
		if(seg.flags.rst){
			free_p(bpp);
			return;
		}
		if(seg.flags.ack){
			free_p(bpp);
			reset(ip,&seg);
			return;
		}
		if(seg.flags.syn){
			/* (Security check is bypassed) */
			/* page 66 */
			proc_syn(tcb,ip->tos,&seg);
			send_syn(tcb);
			settcpstate(tcb,TCP_SYN_RECEIVED);		
			if(length != 0 || seg.flags.fin) {
				/* Continue processing if there's more */
				break;
			}
			tcp_output(tcb);
		}
		free_p(bpp);	/* Unlikely to get here directly */
		return;
	case TCP_SYN_SENT:
		if(seg.flags.ack){
			if(!seq_within(seg.ack,tcb->iss+1,tcb->snd.nxt)){
				free_p(bpp);
				reset(ip,&seg);
				return;
			}
		}
		if(seg.flags.rst){	/* p 67 */
			if(seg.flags.ack){
				/* The ack must be acceptable since we just checked it.
				 * This is how the remote side refuses connect requests.
				 */
				close_self(tcb,RESET);
			}
			free_p(bpp);
			return;
		}
		/* (Security check skipped here) */
#ifdef	PREC_CHECK	/* Turned off for compatibility with BSD */
		/* Check incoming precedence; it must match if there's an ACK */
		if(seg.flags.ack && PREC(ip->tos) != PREC(tcb->tos)){
			free_p(bpp);
			reset(ip,&seg);
			return;
		}
#endif
		if(seg.flags.syn){
			proc_syn(tcb,ip->tos,&seg);
			if(seg.flags.ack){
				/* Our SYN has been acked, otherwise the ACK
				 * wouldn't have been valid.
				 */
				update(tcb,&seg,length);
				settcpstate(tcb,TCP_ESTABLISHED);
			} else {
				settcpstate(tcb,TCP_SYN_RECEIVED);
			}
			if(length != 0 || seg.flags.fin) {
				break;		/* Continue processing if there's more */
			}
			tcp_output(tcb);
		} else {
			free_p(bpp);	/* Ignore if neither SYN or RST is set */
		}
		return;
	}
	/* We reach this point directly in any synchronized state. Note that
	 * if we fell through from LISTEN or SYN_SENT processing because of a
	 * data-bearing SYN, window trimming and sequence testing "cannot fail".
	 *
	 * Begin by trimming segment to fit receive window.
	 */
	if(trim(tcb,&seg,bpp,&length) == -1){
		/* Segment is unacceptable */
		if(!seg.flags.rst){	/* NEVER answer RSTs */
			/* In SYN_RECEIVED state, answer a retransmitted SYN 
			 * with a retransmitted SYN/ACK.
			 */
			if(tcb->state == TCP_SYN_RECEIVED)
				tcb->snd.ptr = tcb->snd.una;
			tcb->flags.force = 1;
			tcp_output(tcb);
		}
		return;
	}
	/* If segment isn't the next one expected, and there's data
	 * or flags associated with it, put it on the resequencing
	 * queue and remind ourselves to ACK it. Then strip off
	 * the SYN/data/FIN and continue to process the ACK (or RST)
	 */
	if(seg.seq != tcb->rcv.nxt
	 && (length != 0 || seg.flags.syn || seg.flags.fin)){
		add_reseq(tcb,ip->tos,&seg,bpp,length);
		if(seg.flags.ack && !seg.flags.rst)
			tcb->flags.force = 1;
		seg.flags.syn = seg.flags.fin = 0;
		length = 0;
	}
	/* This loop first processes the current segment, and then
	 * repeats if it can process the resequencing queue.
	 */
	for(;;){
		/* We reach this point with an acceptable segment; data and flags
		 * (if any) are in the window, and if there's data, syn or fin, 
		 * the starting sequence number equals rcv.nxt
		 * (p. 70)
		 */	
		if(seg.flags.rst){
			if(tcb->state == TCP_SYN_RECEIVED
			 && !tcb->flags.clone && !tcb->flags.active){
				/* Go back to listen state only if this was
				 * not a cloned or active server TCB
				 */
				settcpstate(tcb,TCP_LISTEN);
			} else {
				close_self(tcb,RESET);
			}
			free_p(bpp);
			return;
		}
		/* (Security check skipped here) p. 71 */
#ifdef	PREC_CHECK
		/* Check for precedence mismatch */
		if(PREC(ip->tos) != PREC(tcb->tos)){
			free_p(bpp);
			reset(ip,&seg);
			return;
		}
#endif
		/* Check for erroneous extra SYN */
		if(seg.flags.syn){
			free_p(bpp);
			reset(ip,&seg);
			return;
		}
		/* Update timestamp field */
		if(seg.flags.tstamp
		 && seq_within(tcb->last_ack_sent,seg.seq,seg.seq+length))
			tcb->ts_recent = seg.tsval;
		/* Check ack field p. 72 */
		if(!seg.flags.ack){
			free_p(bpp);	/* All segments after synchronization must have ACK */
			return;
		}
		/* Process ACK */
		switch(tcb->state){
		case TCP_SYN_RECEIVED:
			if(seq_within(seg.ack,tcb->snd.una+1,tcb->snd.nxt)){
				update(tcb,&seg,length);
				settcpstate(tcb,TCP_ESTABLISHED);
			} else {
				free_p(bpp);
				reset(ip,&seg);
				return;
			}
			break;
		case TCP_ESTABLISHED:
		case TCP_CLOSE_WAIT:
		case TCP_FINWAIT2:
			update(tcb,&seg,length);
			break;
		case TCP_FINWAIT1:	/* p. 73 */
			update(tcb,&seg,length);
			if(tcb->sndcnt == 0){
				/* Our FIN is acknowledged */
				settcpstate(tcb,TCP_FINWAIT2);
			}
			break;
		case TCP_CLOSING:
			update(tcb,&seg,length);
			if(tcb->sndcnt == 0){
				/* Our FIN is acknowledged */
				settcpstate(tcb,TCP_TIME_WAIT);
				set_timer(&tcb->timer,MSL2*1000L);
				start_timer(&tcb->timer);
			}
			break;
		case TCP_LAST_ACK:
			update(tcb,&seg,length);
			if(tcb->sndcnt == 0){
				/* Our FIN is acknowledged, close connection */
				close_self(tcb,NORMAL);
				return;
			}			
			break;
		case TCP_TIME_WAIT:
			start_timer(&tcb->timer);
			break;
		}

		/* (URGent bit processing skipped here) */

		/* Process the segment text, if any, beginning at rcv.nxt (p. 74) */
		if(length != 0){
			switch(tcb->state){
			case TCP_SYN_RECEIVED:
			case TCP_ESTABLISHED:
			case TCP_FINWAIT1:
			case TCP_FINWAIT2:
				/* Place on receive queue */
				t = msclock();
				if(t > tcb->lastrx){
					tcb->rxbw = 1000L*length / (t - tcb->lastrx);
					tcb->lastrx = t;
				}
				append(&tcb->rcvq,bpp);
				tcb->rcvcnt += length;
				tcb->rcv.nxt += length;
				tcb->rcv.wnd -= length;
				tcb->flags.force = 1;
				/* Notify user */
				if(tcb->r_upcall)
					(*tcb->r_upcall)(tcb,tcb->rcvcnt);
				break;
			default:
				/* Ignore segment text */
				free_p(bpp);
				break;
			}
		}
		/* process FIN bit (p 75) */
		if(seg.flags.fin){
			tcb->flags.force = 1;	/* Always respond with an ACK */

			switch(tcb->state){
			case TCP_SYN_RECEIVED:
			case TCP_ESTABLISHED:
				tcb->rcv.nxt++;
				settcpstate(tcb,TCP_CLOSE_WAIT);
				break;
			case TCP_FINWAIT1:
				tcb->rcv.nxt++;
				if(tcb->sndcnt == 0){
					/* Our FIN has been acked; bypass TCP_CLOSING state */
					settcpstate(tcb,TCP_TIME_WAIT);
					set_timer(&tcb->timer,MSL2*1000L);
					start_timer(&tcb->timer);
				} else {
					settcpstate(tcb,TCP_CLOSING);
				}
				break;
			case TCP_FINWAIT2:
				tcb->rcv.nxt++;
				settcpstate(tcb,TCP_TIME_WAIT);
				set_timer(&tcb->timer,MSL2*1000L);
				start_timer(&tcb->timer);
				break;
			case TCP_CLOSE_WAIT:
			case TCP_CLOSING:
			case TCP_LAST_ACK:
				break;		/* Ignore */
			case TCP_TIME_WAIT:	/* p 76 */
				start_timer(&tcb->timer);
				break;
			}
			/* Call the client again so he can see EOF */
			if(tcb->r_upcall)
				(*tcb->r_upcall)(tcb,tcb->rcvcnt);
		}
		/* Scan the resequencing queue, looking for a segment we can handle,
		 * and freeing all those that are now obsolete.
		 */
		while(tcb->reseq != NULL && seq_ge(tcb->rcv.nxt,tcb->reseq->seg.seq)){
			get_reseq(tcb,&ip->tos,&seg,bpp,&length);
			if(trim(tcb,&seg,bpp,&length) == 0)
				goto gotone;
			/* Segment is an old one; trim has freed it */
		}
		break;
gotone:	;
	}
	tcp_output(tcb);	/* Send any necessary ack */
}

/* Process an incoming ICMP response */
void
tcp_icmp(
int32 icsource,			/* Sender of ICMP message (not used) */
int32 source,			/* Original IP datagram source (i.e. us) */
int32 dest,			/* Original IP datagram dest (i.e., them) */
uint8 type,			/* ICMP error codes */
uint8 code,
struct mbuf **bpp		/* First 8 bytes of TCP header */
){
	struct tcp seg;
	struct connection conn;
	register struct tcb *tcb;

	/* Extract the socket info from the returned TCP header fragment
	 * Note that since this is a datagram we sent, the source fields
	 * refer to the local side.
	 */
	ntohtcp(&seg,bpp);
	conn.local.port = seg.source;
	conn.remote.port = seg.dest;
	conn.local.address = source;
	conn.remote.address = dest;
	if((tcb = lookup_tcb(&conn)) == NULL)
		return;	/* Unknown connection, ignore */

	/* Verify that the sequence number in the returned segment corresponds
	 * to something currently unacknowledged. If not, it can safely
	 * be ignored.
	 */
	if(!seq_within(seg.seq,tcb->snd.una,tcb->snd.nxt))
		return;

	/* Destination Unreachable and Time Exceeded messages never kill a
	 * connection; the info is merely saved for future reference.
	 */
	switch(type){
	case ICMP_DEST_UNREACH:
	case ICMP_TIME_EXCEED:
		tcb->type = type;
		tcb->code = code;
		tcb->unreach++;
		break;
	case ICMP_QUENCH:
		/* Source quench; reduce slowstart threshold to half
		 * current window and restart slowstart
		 */
		tcb->ssthresh = tcb->cwind / 2;
		tcb->ssthresh = max(tcb->ssthresh,tcb->mss);
		/* Shrink congestion window to 1 packet */
		tcb->cwind = tcb->mss;
		tcb->quench++;
		break;
	}
}
/* Send an acceptable reset (RST) response for this segment
 * The RST reply is composed in place on the input segment
 */
void
reset(ip,seg)
struct ip *ip;			/* Offending IP header */
register struct tcp *seg;	/* Offending TCP header */
{
	struct mbuf *hbp;
	uint16 tmp;

	if(seg->flags.rst)
		return;	/* Never send an RST in response to an RST */