Re: [lwip-users] TCP: ACK problem?

[David Empson]

Andre replied to me directly. I've explained to him how to join the 
mailing list and do this properly, but in the meantime, I've attached a 
copy of his lwipopts.h and my subsequent reply follows, in the hope that 
others can add their comments on the problem.


A second opinion from someone more familiar with the LWIP memory 
allocator (we use a custom one so I haven't looked at LWIP's one in 
detail) and your particular target's network driver would be helpful, 
but from a quick look I think my theory about running out of pbufs is on 
the right track.

Your PBUF_POOL_SIZE is only 6. Depending on the implementation of your 
network driver and application, this pool may be used to store both 
received packets and packets which are copied for transmission by 
tcp_write().

With only 6 pbufs available for allocation and 3 of them occupied by 
unacknowledged outgoing packets, you would only need 3 more allocated 
anywhere else in the system to have none left for processing receive 
packets, resulting in your network driver discarding all incoming 
packets rather than being able to deliver them to ip_input().

As I alluded to in my first message, you should be able to confirm this 
by looking at the data generated by LWIP_STATS and see if you are 
getting memory allocation errors for the MEMP_PBUF_POOL type of memory 
pool, which go up each time another packet is received if I'm right. To 
see the error counting logic, follow the code through from 
pbuf.c:pbuf_alloc() to memp.c:memp_malloc(), with reference to stats.h.

If that error is confirmed, you will need to identify where the other 
three pbufs have got to and/or limit the number of consecutive calls you 
make to tcp_write() whiledata has been acknowledged and/or increase 
PBUF_POOL_SIZE (which would result in lwip being allocated more memory 
overall), or come up with some other strategy to limit the conflicting 
memory allocation.

On 2013-08-21 20:54, André P. wrote:
> Hi David,
>
> Thank you for your fast response. I have attached my lwipopts.h file.

On 2013-08-21 14:46, David Empson wrote:
> On 2013-08-21 03:45, André P. wrote:
> > Hi,
> >
> > I have an application where I stream data regularly from an LPC board to
> > a laptop. This works fine, until after some time the transmitter sends
> > two packets without waiting for an ACK, and the receiver almost
> > immediately acknowledges both using one packet. Somehow, the LWIP stack
> > does not accept this ACK, and keeps resending a package.  I have attached
> > the Flow Graph. Why does LWIP not accept the ACK message 166213? It
> > keeps resending Seq 162622.
> > I am using LWIP 1.4.0. Thank you for your support.
>
> The pattern looks like LWIP stopped processing the received ACK packets
> after the one with Ack=162622. LWIP subsequently sent three packets with
> new data (162622, 163648 and 165108), then timed out waiting for the ack
> for the oldest unacknowledged packet (162622), resending it with slowing
> retries: 1 second delay, then 3, 6, 12, 24, 48; all the subsequent ACKs
> were not processed.
>
> You'll probably need to look at LWIP's internal error counts to confirm
> this, but my guess is that you ran out of pbufs, due to having three
> outgoing packets occupying all remaining pbufs, and none are left to
> store the received ACKs. If so, you need to allocate more pbufs and/or
> limit the amount of transmit data you can have outstanding to ensure
> there will always be free buffers for received packets.
>
> A copy of your lwipopts.h might help to analyse the problem.
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/lwip-users

lwipopts.h
Description: Text document