[PATCH] : libc_r/uthread/uthread_write.c

[Dan Langille]

On 18 Sep 2003 at 7:50, Daniel Eischen wrote:

> On Wed, 17 Sep 2003, Dan Langille wrote:
> 
> > On 16 Sep 2003 at 20:49, Daniel Eischen wrote:
> > 
> > > On Tue, 16 Sep 2003, Dan Langille wrote:
> > > 
> > > > I've had preliminary success with this patch.  More testing needs 
> > > > to be done, but in the meantime, I would appreciate reviews and 
> > > > comments.  The patched code is available from
> > > > http://beta.freebsddiary.org/tmp/uthread_write.c and the patch
> > > > appears below.
> > > > 
> > > > In short, the logic has been changed to ensure that if __sys_write
> > > > returns zero, this value is returned by _write.
> > > 
> > > I think this is not quite correct.  Since libc_r is looping
> > > and some data may have been read, then the partial byte count
> > > should be returned, not zero.  It is possible the partial byte
> > > count could also be zero in some cases, so it would result
> > > in zero being returned in those instances.
> > 
> > I see what you mean. Please have a look at 
> > http://beta.freebsddiary.org/tmp/uthread_write.c2
> > The patch appears at the end of this message.
> 
> Right, this seems correct to me.

All our testing on this patch has been successful.  I'm going to do a 
few more tests on different hardware under 4.8-stable.

What's the next step?  Commit it?  Get others to test with it first?  

> > > I think the problem lies with the SCSI tape device. It should
> > > either return 0 or -1 with errno=ENOSPC on a write that detects
> > > an EOT, not partial byte count.
> > 
> > You are referring to sa(4)?
> 
> Yes.
> 
> > > If you are using libkse or
> > > libthr, you will get a partial byte count and not zero because
> > > the tape driver returns the (partial) bytes written.  So exiting
> > > the loop in libc_r and returning 0 would only seem to correct
> > > the "problem" for libc_r.
> 
> If there is a difference, it could be because libc_r is using non-blocking
> IO behind the scenes, and sa(4) may be returning partial byte count
> in the non-blocking case and 0 (or -1 and ENOSPC) in the blocking case
> (which is what you'd get using libkse/libthr).
> 
> > The problem  found when running under pthreads on 4.8-stable [i.e. 
> > EOT is not returned to the application code] is not found with libkse 
> > on 5.1-current.

FWIW: our regression tests are failing under 5.1 and we suspect that 
MTIOCERRSTAT ioctl() has changed since 4.8.  We're getting:

btape: dev.c:1119 Doing MTIOCERRSTAT errno=22 ERR=Invalid argument

We'll continue with our 5.1 work, but we'd like to finish up with 4.8 
ASAP.
-- 
Dan Langille : http://www.langille.org/