linux readahead bug is back
alan at lxorguk.ukuu.org.uk
Fri Nov 23 10:46:56 UTC 2007
On Fri, 23 Nov 2007 02:04:03 -0500 (EST)
"D. Hugh Redelmeier" <hugh at mimosa.com> wrote:
> | From andre at bwh.harvard.edu Fri Nov 23 00:33:13 2007
> | My father just bought a F8 DVD, but it failed the mediacheck. It
> | turns out that the vendor failed to apply zero-padding to the ISO file
> | before burning, and that the failure occurs about 40 KB before the end of
> | the disc, presumably due to the readahead bug having returned. This had
> | been fixed when F7 was released, but apparently is back now in both the
> | updated F7 and F8 kernels. Anyone know when this regression happened, and
> | if there are plans to fix it?
> My theory, based on a few samples, is that the "readahead bug" was
> never fixed. It only manifests itself on some drives. Older drives,
> I think.
It was fixed at least for all the drives we could test. Fedora 7 and 8
use libata which uses the SCSI layer for ATAPI. The SCSI layer correctly
understands partially successful write returns.
If you could file a bug and include the info on the drive and firmware
that would be handy.
The old IDE layer had problems when reading disks raw it would get near
the end (which isn't precisely defined for CD-R/CD-RW etc). When it asked
for 64K and got back 32K it couldn't cope. The SCSI layer based libata is
quite happy in that situation and will only error if the user then tries
to read beyond the 32K (or whatever) it actually got.
More information about the fedora-list