[lwip-users] [lwip] pcb->snd_nxt or pcb->snd_max?

[H.J.W.]

        Hi!
        I am trying to port lwip to a wireless environment without an OS. So I 
simulate a high lost rate (25 will lost) network enviroment in PC and 
experienced a lot of problems, most are retransmission related. Today I got a 
'RST' segment from the peer in 'FIN_WAIT_2' state when trying to close a 
connection. And I think I have found out the reason.
        
        Here is a code piece from function "tcp_process":

        case LAST_ACK:
    tcp_receive(pcb);
    if(flags & TCP_ACK && ackno == pcb->snd_nxt) {
      DEBUGF(DEMO_DEBUG, ("TCP connection closed d -> d.\n", inseg.tcphdr->src, 
                inseg.tcphdr->dest));
      pcb->state = CLOSED;
      pcb->flags |= TF_CLOSED;
    }


        I believe it is safer to use 'pcb->snd_max' than 'pcb->snd_nxt' in 
several places of function "tcp_process", because 'pcb->snd_nxt' may be
changed when "tcp_slowtmr" was called. For example, when the pcb send a 'FIN' 
to destination,
but the segment was lost in the network, so the 'FIN' segment was placed in the 
unacked queue together with other unacked segments, then "tcp_slowtmr" was 
called, all unacked segments are moved to unsent queue exept the first unacked 
segment, the 'pcb->snd_nxt' was changed to the seq number of the first unsent 
segment, the first unacked segment was retransmited and was received by 
destination. Then, a segment from destination with the ack number equal to 
'pcb->snd_nxt' comes, well, what will happen in "tcp_process"? The pcb will be 
purged and the 'FIN' segment will never transmit to destination! 

        And I really suggest to test lwip in tough network environment, for 
example, 25% lost rate and 10 segments per second with only 32k byte data ram.


        Regards!


        
        
¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡H.J.W.
address@hidden
¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡¡2002-11-11

[This message was sent through the lwip discussion list.]