Re: Deadlock between GEOM and devfs device destroy and process exit.

On Mon, Feb 01, 2010 at 12:01:13PM +0200, Alexander Motin wrote:
> Pawel Jakub Dawidek wrote:
> > On Sat, Jan 30, 2010 at 12:44:51PM +0100, Pawel Jakub Dawidek wrote:
> >> Maybe I'll add how I understand what's going on:
> >>
> >> GEOM calls destroy_dev() while holding the topology lock.
> >>
> >> Destroy_dev() wants to destroy device, but can't because there are
> >> threads that still have it open.
> >>
> >> The threads can't close it, because to close it they need the topology
> >> lock.
> >>
> >> The deadlock is quite obvious, IMHO.
> > 
> > Guys, changing destroy_dev() to destroy_dev_sched() in geom_dev.c fixes
> > the problem for me (at least it makes race window so small that I can't
> > reproduce it). Is there anyone who isn't happy with such a change?
> 
> Have you done some locking there? Because my system crashes with such
> straightforward change, when g_dev_close() called after geom node being
> destroyed. Attached patch fixes that for me, but I have doubts that it
> is complete. There is still seems to be a race with new I/O requests and
> ioctl's, that are not protected by topology lock. At least if devfs code
> doesn't handle it somehow.

Devfs prevents new threads from entering cdevsw methods after
destroy_dev_sched() is called. It is driver responsibility to take care
about other code pathes that may access cdev.