SV: [lwip-users] TCP retransmit bug

[Andreas Stenius]

Hi Karl,

I'm no expert at lwIP, but I did a quick lookup in the code and came up with
this.

> 4. calls tcp_output()

This will result in pcb->unacked to be filled again, after it has been
transmitted.

> But if further timeouts occur, and no acks have been received, both
> tcp_slowtmr() and tcp_rexmit() find that pcb->unacked is null, and
> therefore no subsequent retransmission is attempted.

Not true, since the pcb->unacked is != NULL, unless we've got our ack, and
no retransmission
should be done.

> Shouldn't both of these functions be modifed to check for segments
> on pcb->unsent in addition to pcb->unacked?

The segments in the unsent queue are not up for retransmission, since
they're still
waiting to be sent. The slowtmr does some polling (don't know what happens
here),
and if it returns ERR_OK, calls tcp_output, which will take care of the
segments in the unsent queue.


//Andreas