Hi Herbert,

Herbert Xu wrote:

>Firstly there is a sock leak there.  The following patch fixes it.
>The white space damage is not mine :)
>
>Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
>  
>

you are right, I missed that, but I prefer the patches attached to this 
email. Now if netlink_broadcast_deliver function delivers correctly the 
packet, you decrease sock refcount and function returns 1. I think that 
I got confused because netlink_broadcast_deliver returns 0/-1.

>Secondly, I'm dubious about the patch as a whole.  For instance, what
>exactly is the wake_up_process() bit trying to do? Surely that process
>would've been woken up multiple times already if its queue is full.
>

This is what I theorically expected, but in practice if you stress a 
netlink socket sending a big bunch of information in a short period of 
time from kernel space to user space, socket overruns easily. That's why 
I wake up the user process to make it process information stored in the 
queue. Socket doesn't overrun anymore with my patch.

>And what is it going to do with thread groups?
>  
>

currently broadcast sockets can still overrun. I have a set of patches 
that I'll submit as soon as I test them and I finish my boring exams.

regards,
Pablo