On Mon, Mar 25, 2002 at 10:50:17AM +0100, Axel Kittenberger wrote:
> 
> > This is me talking prior to having coffee, but Chapter 3 of the
> > Rubini/Corbet book says:
> >
> >   The flush operation is invoked when a process closes its copy of a file
> >   descriptor for a device; it should execute (and wait for) any outstanding
> >   operations on the device. This must not be confused with the fsync
> > operation requested by user programs. Currently, flush is used only in the
> > network file system (NFS) code. If flush is NULL, it is simply not invoked.
> >
> > I guess it doesn't specifically say it's not called in midstream, but
> > it reads as if flush() is called on /only/ close(). I may test this
> > today, just for fun.
> 
> Oh thats interesting, indeed, so the function name "flush" is just 
> contra-intentional. Oxay I know now how I could have written this driver 
> without patching the kernel.....
> 

And FWIW, I tested this behavior with a dummy chardev and printks()
around open(), release, and flush(). flush() is indeed called only on
each close.

> Still the basic issue/idea is remaining. release() is defined as int return 
> type, but everywhere it's called it's value is discarded. (except internally 
> in "intermezzo" whatever that is)
> 
> btw: blkdev_put()  has int return type and seems to return correctly the 
> return value from release()s for block devices, so I guess it would be the 
> right thing for char devs to do also.
> 
> The other way I would also see as okay is to state release() can't return 
> anything senseful to anybody, bet then declare it as void return instead. But 
> as the state is currently it's suboptimal from both views.

Agreed, but the question is which approach to use. :) Declaring it as void
sounds like it may involve a lot of driver fixup work.


Regards,
- Dave
  dave@codewhore.org