Reattach/redetect allways connected umass device - is it possible ?

[Bernd Walter]

On Mon, Mar 28, 2005 at 03:32:31PM +0200, Poul-Henning Kamp wrote:
> In message <1112015844.1022.44.camel at localhost>, Vladimir Grebenschikov writes:
> >÷ ÐÎ, 28/03/2005 × 15:04 +0200, Bernd Walter ÐÉÛÅÔ:
> >> On Mon, Mar 28, 2005 at 04:58:31PM +0400, Vladimir Grebenschikov wrote:
> >> > ? ??, 28/03/2005 ? 14:13 +0200, Poul-Henning Kamp ?????:
> >> > > In message <20050328114633.GZ14532 at cicely12.cicely.de>, Bernd Walter writes:
> >> > > 
> >> > > >> camcontrol detach da0; camcontrol rescan all 
> >> > > >> helps, but, it should be much better if it will be issued automatically.
> >> > > >
> >> > > >Yes - GEOM seems to ignore media change signals from drives.
> >> > > >I've added PHK to the recipient list - maybe he has an idea about this
> >> > > >problem.
> >> > > 
> >> > > No, GEOM doesn't ignore any such thing, because as far as I know
> >> > > GEOM doesn't get any such thing to ignore in the first place.
> >> > 
> >> > So, let's imagine following situation:
> >> > 
> >> > We get SCSI BUS with removable da device.
> >> > device detected as da0 and not mounted.
> >> > Device disconnected from SCSI bus.
> >> > And finally, another device with different geometry connected with same
> >> > SCIS ID.
> >> 
> >> This ist not a *media* exchange - this is a *device* and in
> >> this case even a scbus exchange.
> >
> >Ok, so my case is media exchange, not device exchange.
> >How it is supposed to work in this case ? 
> 
> That's a very good question.

He disconnects and reattaches the whole device.
The disconnect blocks somewhere until something magic happens and then
he gets the new device.
Under USB you have to reprobe the device, as a freshly attached device
is completely unconfigured.
There is absolutely no chance that you can reconnect a USB device
without getting a new instance.

> The original intent in GEOM was that the geom instance would represent
> the media while the drive (if separate for the media) would have
> another access mechanism.

Sounds reasonable.

> Driver support for this is not really meaterialized and therefore
> the model now is that when the media is ejected the geom device
> is removed and a new one created right away, even if a new media
> is not inserted right away.

Do you get eject notifications from any device type?

> This works with the broken-by-design CDrom ioctls etc.
> 
> What is missing is to tickle GEOM when the media is inserted so
> that the tasting takes place.

Mmm.
One of my problems is that it is possible with SCSI MO drives to
eject the media, reinsert a new one and errously continue to work
with the old GEOM layout.
I already trashed the data on one media by forgetting to trigger
a reprobe - since then I use a mount script.

-- 
B.Walter                   BWCT                http://www.bwct.de
bernd at bwct.de                                  info at bwct.de