On 07.09.2010, at 11:40, Bernhard Kohl wrote: > Am 07.09.2010 04:04, schrieb ext Christoph Hellwig: >> On Mon, Sep 06, 2010 at 05:39:00PM +0200, Alexander Graf wrote: >> >>> On 06.09.2010, at 17:06, Bernhard Kohl wrote: >>> >>> >>>> From: Bernhard Kohl<bernhard.k...@gmx.net> >>>> >>>> Allow symbolic links which point to /dev/sgX devices. >>>> >>> Couldn't you send an SG_IO test ioctl over and see if it works? I really >>> dislike the whole file name magic matching. >>> >> You could, but the result would not be what you expect, given that every >> /dev/sd* device and more in Linux support it. >> >> What we really need is to stop shoe-horning scsi pass through support >> into the block layer. Once we finally get our generic thread offload >> facilily there is no need for it anymore at all, scsi-generic can >> simplify offload the SG_IO ioctl and be done with it. I even have an >> old prototype for this, just waiting for the generic thread offload >> to get merged before resurrecting it. >> > > Wouldn't it be OK to apply this patch in the meantime before we get > this complete new solution? I need to handle about 15 /dev/sg* disks > which are randomly numbered. With this patch I can use the udev > generated links in /dev/disk/by*, which makes life much easier. >
I agree. I'd take this one in for now, as it improves the situation and then when all world blossoms we can take Christoph's game-changing framework :). Kevin, what do you think? Alex
