In message <20031015101141.727cfa07.ggm@apnic.net>, George Michaelson writes:
>
>I removed a USB 2.0 cardbus card, with device still attached by cable. the dis
>k
>had not been mounted in user FS space, but had been probed and attached to SCS
>I
>via atapibus/umass mappings. this caused a panic. my memory tells me that even
>on 4.2 on vaxen, arbitrary removal of disks was sometimes a bit painful for a
>kernel. Is this still true?
>
>Should I have detached the sd* device by removing the USB cable first? Is ther
>e
>some userspace trick which can make it 'safe' to yank a controller with attach
>ed
>devices and not panic a kernel?
>
>I have a vestigial fear that I can't remove the cardbus card if I've ever
>mounted a disk via that card, which is a bit different. (I remember seeing thi
>s
>also panic some time ago this year)
>

I raised a similar issue quite a while ago -- my problem was a PCMCIA 
SCSI controller.  The answer at the time was that since there's no way 
to speak of a controller directly at user level, there was no way to 
"close" it, and hence no way to remove the card safely.  I don't recall 
seeing any fixes that address it, but since I haven't used the card in 
a long time I haven't worried about it.


		--Steve Bellovin, http://www.research.att.com/~smb