cvs commit: src/sys/dev/ata atapi-cd.c atapi-cd.h

Nate Lawson nate at
Fri Nov 16 14:32:49 PST 2007

Poul-Henning Kamp wrote:
> In message <473DCF61.6010107 at>, Scott Long writes:
>>> What I've sortof talked phk into is that if I return the "right" error 
>>> code in the "start" function, it will back off and not retry nor chatter.
>>> We still need the "access" function to return OK if HW present or we 
>>> cannot send ioctl calls to an empty drive.
>> IMHO, overloading g_start just adds more cruft to the problem.
> The fundamental problem is that the CDROM ioctls don't distinguish
> between drive and media.
> In a nutshell: we are forced to let the open succeed, so the ioctl
> to close the tray can be issued, before we can get at the media.
> The problem with letting the open (aka g_access) succeed, is that
> the GEOM tasting starts.
> Redefining the ioctls is not a realistic option at this point, so
> the task is to find a workaround for the API-deficiency.
> What sos@ and I talked about is that we need to have an error return
> that says "The drive is here, but there is no media", to fail the
> tasting without GEOM reporting I/O errors on the console.
> The simplest way to do this, is probably to overload ENOENT or some
> other currently undefined errno for this effect.

Well, there are two questions:  media present (yes/no) and drive capable
of telling if media present without just trying to read it (yes/no).
For the latter question, a CDROM is yes and a floppy drive is no.  For
the latter, you need a third return code which is "I don't know, and
only attempt a read if the user explicitly asked for it with a setting".

Since boot-time automatic tasting is only needed if you're going to
automount it, I doubt it's a huge inconvenience for users that want to
automount a floppy to incur the couple seconds delay while it is probed.
 For the rest of us, we save boot time.

Note that this wouldn't change the fact that tasting would occur
synchronously once someone asked to mount a floppy.  So any problem with
making this change?


More information about the cvs-src mailing list