GEOM_VOL panics when audio CD is present

[Poul-Henning Kamp]

In message <20041230121650.GB784 at darkness.comp.waw.pl>, Pawel Jakub Dawidek wri
tes:
>
>--MfFXiAuoTsnnDAfZ
>Content-Type: text/plain; charset=iso-8859-2
>Content-Disposition: inline
>Content-Transfer-Encoding: quoted-printable
>
>On Wed, Dec 29, 2004 at 05:28:09PM +0100, Pawel Jakub Dawidek wrote:
>+> On Wed, Dec 29, 2004 at 04:53:05PM +0100, Dag-Erling Sm?rgrav wrote:
>+> +> g_vol_ffs_taste() tries to taste audio CDs, triggering a KASSERT in
>+> +> g_io_request() because 65536 is not a valid offset on a medium with
>+> +> 2352-byte sectors.  The exact panic message is "wrong offset 65536 for
>+> +> sectorsize 2352" (sys/geom/geom_io.c lines 250-251).
>+> +>=20
>+> +> None of this code has been touched recently, so I suspect the same bug
>+> +> is present in 5.3, though I haven't had a chance to test it yet.
>+>=20
>+> Those KASSERT()s are hiden under DIAGNOSTIC in RELENG_5.
>+>=20
>+> Anyway, I had simlar problems. I'll prepare a patch to try soon.
>
>Hmm, I'm not sure what is correct way to fix it.

The correct fix is of course to fix the bug in the geom_vol_ffs
class that makes it even try to read things from providers which
have sectorsizes it cannot cope with.

>Maybe we should just remove those KASSERTs?

The KASSERTS are correct and should not be removed.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.