Re: kern/2727 (Writing to SCSI tape panics system with 'done < 0; strategy broken message)

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Sep 02, 2008 at 09:39:47PM +0200, Manuel Bouyer wrote:
> On Tue, Sep 02, 2008 at 07:15:23AM +0000, David Holland wrote:
> > On Tue, Sep 02, 2008 at 07:07:02AM +0000, dholland%NetBSD.org@localhost 
> > wrote:
> >  > State-Changed-From-To: suspended->open
> >  > While we're looking at st(4)...
> > 
> > So... here, I think the problem with Manuel's 2002 patch (which has
> > been committed at some point) is that it wasn't meant to deal with an
> > error condition. But I note that st.c now has this in stdone:
> > 
> >     bp->b_error = error;
> >     bp->b_resid = xs->resid;
> >     /*
> >      * buggy device ? A SDLT320 can report an info
> >      * field of 0x3de8000 on a Media Error/Write Error
> >      * for this CBD: 0x0a 00 00 80 00 00
> >      */
> >     if (bp->b_resid > bp->b_bcount || bp->b_resid < 0)
> >             bp->b_resid = bp->b_bcount;
> > 
> > which will both avoid a panic and also prevent upper-layer code from
> > thinking the write went through.
> > 
> > So I think this PR can be closed. Manuel? Anyone else? Is the
> 
> Not yet, there's still a path that can lead to failure. I've debugged
> this with a user running into it and have some patch. I need to clean it up
> and request feedback on tech-kern.

On second though this bug can be closed. The issue I've debugged recently
is a bit different from this one (done > 0  with b_error set), and gives
a different panic.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--