Re: [lwip-devel] memory leak with non blocking send on lwip 1.4.1.

[M B]

Hi,

We have had serious issues with do_writemore() in 1.4.1 on non-blocking 
sockets. I have had the intention to look deeper into it, but time never 
allows me to.. Our quick-fix was to revert the whole function back to 
1.4.0 to get a long term reliable communication.

In our case, we are using about 10 non blocking sockets in a very low on 
ram system. Part of the problem was related to memory if *my* memory 
serves me right :).

This is probably not related to your issues, but it might be an easy 
test to do - just copy do_writemore() as it is from 1.4.0.

 Micael

On 06/21/13 11:33, RAc wrote:
> Hi there,
>
> sorry if this should be a dupe; I did check the Forum and to my best
> knowledge this hasn't been raised before - apologies if I should have
> overlooked something...
>
> we are using the lwip socket API to communicate. This works fine except that
> the lwip_write() function occassionally hangs; we could trace this scenario
> to a connection being closed at the "wrong" time. Not a problem by itself; a
> solid implementation will avoid blocking scenarios anyways.
>
> It turns out that lwip_write translates to
>
> lwip_send(s, data, size, 0);
>
> since a valid parameter for the flags field is MSG_DONTWAIT, we replaced the
> calls to write() (which is a macro wrapper for lwip_write()) with a call to
>
> lwip_send(s, data, size, MSG_DONTWAIT);
>
> and consequently evaluate all the returns. This works fine and doesn't cause
> hangs anymore; however, there is a memory leak now. The closest I came to
> debugging it is that the memory section not freed appears to be a pcb
> allcated in tcp_alloc in tcp.c:
>
>    pcb = (struct tcp_pcb *)memp_malloc(MEMP_TCP_PCB);
>
> I've tried to further follow the scenarios where the pcb normally is
> deallocated, but this goes a little bit too far into the guts of lwip for me
> to pinpoint further without a lot of digging into the code.
>
> Are there known issues with timeout sends under lwip 1.4.1.?
>
> I also looked into the SO_SNDTIMEO socket option, but a) the declaration in
> sockets.h indicates that the option is unimplemented and b) even though it
> says so, it appears that there is an implementation for it but it seems to
> end up in the same code path as the dontwait.
>
> Any input is appreciated, thanks. Please do NOT argue about the thing being
> unsupported or something the like without at the same time providing some
> option to avoid the hang. :-) Thanks!
>
>
>
>
> --
> View this message in context: 
> http://lwip.100.n7.nabble.com/memory-leak-with-non-blocking-send-on-lwip-1-4-1-tp21577.html
> Sent from the lwip-devel mailing list archive at Nabble.com.
>
> _______________________________________________
> lwip-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/lwip-devel