Bug#690505: broken usb device breaks direct i/o even to non-usb drives

[Jonathan Nieder <jrnieder@gmail.com>]

Hi,

mail user wrote:

> plugging in a fake (128 that later turned out to be just 2Gb) flashdrive
> in the usb port, dd it to /dev/null, ctrl-c, dd zeroes to it, ctrl-c,
> dd unresponsive to ctrl-c

Thanks for a clear report.  That sounds like a serious bug.  Is it
reproducible?

If so, full "dmesg" output after reproducing it and running

	echo w >/proc/sysrq-trigger

would be very interesting.

[...]
> then
> rmmod -f usb-storage
>
> then
> unplug the flashdrive
>
>    * What was the outcome of this action?
>
> forced removal of usb-storage produced the following kernel dmesg
> (ignore the tainted message, it is from the zfs/spl modules):
>
> Oct 15 00:40:50 homerouter kernel: [88756.711046] Disabling lock debugging due to kernel taint
> Oct 15 00:40:50 homerouter kernel: [88756.711092] usbcore: deregistering interface driver usb-storage
> Oct 15 00:40:51 homerouter kernel: [88757.349748] BUG: unable to handle kernel paging request at ffffffffa05b9020
> Oct 15 00:40:51 homerouter kernel: [88757.349831] IP: [<ffffffffa018ef60>] scsi_device_put+0x11/0x3c [scsi_mod]

Yeah, this part doesn't surprise me.  Forced unloading of a module
doesn't generally work very well --- it's a hack to cope with some bad
situations, but generally the best policy is to reboot ASAP
afterwards.

Hope that helps,
Jonathan