Re: IDE CDROM tail read errors

[Bill Davidsen <davidsen@tmr.com>]

Alan Cox wrote:
> On Mer, 2006-03-29 at 21:11 +0200, Krzysztof Halasa wrote:
> > # ls -l
> > -rw-r--r-- 1 root root 687177728 Mar 29 15:51 img-15.iso
> > -rw-r--r-- 1 root root 687177728 Mar 29 15:58 img-15a.iso
> >
> > The files are just truncated FC5d1 images (57344 bytes missing).
>
> The final partial read is dropped rather than partially completed.
>
> > # cat /sys/block/sr0/size
> > 1342264 (i.e., the same as with 2.6.15 + drivers/ide)
> >
> > # cat /dev/cdrw > img-16a.iso
> > cat: /dev/cdrw: Input/output error
> >
> > # cat /sys/block/sr0/size
> > 1342256 (looks like it has been adjusted to .iso image size / 512 when
> >          the first I/O error occured)
>
> The SCSI layer does this bit for everyone. Its actually not libata or
> the PATA drivers that have done the work here. You should find ide-scsi
> does the same.
>
> I patched the old IDE driver a bit to try and deal with this and if you
> want the patch to hack on and tidy up further feel free. 
Any hope in hell of getting this fixed? It's been broken for too long to 
remember. IIRC the problem is that on a read which hits EOF, instead of 
returning partial data read and no error, the status returned is just 
some "didn't work" variant and the partial read is lost. I also seem to 
remember that if the length of the ISO was a multiple of a magic number, 
32k, or the metric hexadecimal phase of the moon, you got a clean EOF.

As you say, ide-scsi fixes this, if the people who don't like it would 
make the alternatives work right it wouldn't be an issue...

Thanks for the info.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/