Re: [tipc-discussion] [PATCH v3] tipc: return POLLOUT for sockets in an unconnected state

[Erik H. <eri...@er...>]

> Looks good, thanks Erik.  Help me gauge the impact of this -- i.e.
> is it a rare corner case, or an easy to trigger bug that is
> impacting a lot of people?

It's quite easy to reproduce if you have an application that 
(implicitly) establishes and tears down connections at high rates.
A link congestion lasting 1 second will cause a lot of connections to 
'hang'.

>   If it has a high impact, we can
> suggest it for "net" and hence 3.7 inclusion, otherwise we
> should not needlessly add to the post merge window burden, and
> instead nominate it for net-next which means 3.8.

Not as high as the packet loss bug, but still, this forces one of our 
TIPC applications to do an extra syscall (explicit connect) to send data 
(read, establish a call/context).

I still think it's fine to suggest for net-next, (we backport these 
critical fixes to older kernels anyway).

//E