From: Jeff Dike <jdike@ka...> - 2002-05-13 21:19:11
> Transferring data to a uml session (tun/tap networked) will often
> stall at the end. A good trigger is doing an scp of a few files, it's
> very easy to trigger. It will appear to hang at the very end,
> initiating other  traffic will cause the transfer to finish.
I never looked at that problem too closely because I never saw it happen.
The fact that it can be rescued with more traffic makes it sound like
an interrupt is getting lost. I'll see if I can make it happen.
From: Jeff Dike <jdike@ka...> - 2002-05-14 14:46:07
> Here's how I can reproduce it with a few tries (doesn't always hang in
> the first go, but 2nd or 3rd it usually does).
Awesome, got it to happen on the third try.
I'll figure out what's going on here. It's not a simple case of a lost
interrupt. Banging on the keyboard, disk IO, and network traffic on a
different interface all fail to wake up the connection.
It has to be traffic on the same interface.