Re: [lwip-users] lwip_read hanging and wrong sequence number

[address@hidden]

Cristina Marcucci wrote:
> 1. My thread reads data with lwip_read (I have set a timeout of 36 seconds
> as a socket option, but the problem occurs also without the timeout
> directive). After working fine for some time, my thread "hangs" on the
> lwip_read, no more data received or a timeout happening. But the incoming
> messages are are still being processed and acked at TCP level all the time.
> My window size of course shrinks more an more, and finally my receive queue
> gets full, proving that my thread is not receiving anylonger. Wireshark
> trace attached.
>    
Can you see where it hangs? If you can debug your target, you should be 
able to tell us exactly where it hangs in lwip_read(). The only thing I 
can imagine hanging there is sys_arch_mbox_fetch(), which is a function 
of the AVR port. If that hangs while the recvmbox is growing, it 
suggests there is a problem with the mbox functions of that port. It 
should at least time out if it can't fetch new data.

Oh, this applies to 1.3.x only, which version of lwIP are you using?
> 2. In a different scenario, after running fine for some time, my application
> sends a wrong sequence number. Wireshark trace attached.
>    
Without knowing your protocol, I'd say what we see here is a lost packet 
rather than a wrong sequence number. Can you check whether that's the 
case somehow? You might include a counter in the sent packets to check 
that with wireshark. If that's the case, someone is dropping packets: 
either lwIP (or the netif driver, rather) or a network component between 
your device and wireshark. The netif driver is the most suspicious here...

What's strange is that for the retransmission, the lost packet is 
skipped again... And even more strange is that the device seems to be 
dead after that (only one retransmission, no ACKs to the retransmissions 
from the .82 host).

Simon