Re: [PATCH 1/2] BKL: Remove BKL from default_llseek()

[Jamie Lokier <jamie@shareable.org>]

Oliver Neukum wrote:
> Am Mittwoch, 18. November 2009 18:27:30 schrieb Jamie Lokier:
> > > No. Your logic is flawed
> > > 
> > > The BKL is protected something here - it protects the change of offset
> > > with respect to other BKL users within drivers. The question is what if
> > > anything in any other driver code depends upon the BKL and uses it to
> > > protect f_pos. Probably very little if anything but a grep for f_pos
> > > through the drivers might not be a bad idea before assuming this. Very
> > > few touch f_pos except in their own llseek method.
> > 
> > Of course, drivers shouldn't be using f_pos outside their llseek
> > method, as they should all behave the same with pread/pwrite as with
> > llseek+read/write.
> 
> Might not a driver update f_pos after read/write?

It could indirectly, through *ppos.  

There should be no direct accesses to f_pos outseek llseek.  If there
are still, those might indicate driver bugs.  (I'm not 100% sure about
this - hence asking).

Drivers used to update f_pos indirectly through *ppos, and for this,
Alan's observation about BKL protecting the value from changing does apply.

But nowadays, even that doesn't happen.  sys_read() and sys_write()
make a copy of f_pos using file_pos_read(), so drivers cannot see the
value change during the call - except for their own change.

I find myself wondering why the VFS isn't responsible for the position
update instead of the driver...  Would it be a valid cleanup to move
it from the driver to VFS?

-- Jamie