On Mon, Oct 14, 2002 at 06:25:51PM +0900, YAMAMOTO Takashi wrote:
> > hi,
> > 
> > the problem with this fix is that it opens a window where another thread
> > could see uninitialized pages in the region of the file that uiomove()
> > failed to write to.  however, unwinding the softdep state at this point
> > is no doubt more hassle than it's worth, so what we need to do is just
> > make sure that the kernel initializes the pages if uiomove() didn't
> > do it.  just zeroing the range that uiomove() would have filled is fine.
> > ideally we could use a fault-tolerant kzero() function (similar to kcopy()),
> > but that doesn't exist.  in practice we'll never fault here because we just
> > allocated the pages right before this, so memset() is good enough for now.
> > 
> > -Chuck
> 
> i see..
> then how about attached one?
> thanks.
> 
> YAMAMOTO Takashi


that look fine.  thanks for working on this!

-Chuck