Package: usbmount
Version: 0.0.10
System configuration:
i386, Debian sarge, Kernel 2.5.15.6 (self compiled), udev 056 (standard
sarge), usbmount 0.0.10 (standard sarge, but same code and bug is also
in 0.0.14 of sid), USB controller UHCI, USB stick SanDisk Cruizer
Micro.
Behaviour:
When inserting an USB stick, it is correctly assigned /dev/sda1 by
udev, and then mounted to /media/usb0 by usbmount, and then linked to
/var/run/usbmount/* by 00_create_model_symlink in the run-parts
directory /etc/usbmount/mount.d/.
When removing the USB stick the /dev/sda1 is removed by udev, but
usbmount then fails to call 00_remove_model_symlink (or anything else)
in the second run-parts directory /etc/usbmount/umount.d/.
Bug Location:
What goes wrong is the script /usr/share/usbmount/usbmount, which in
its section for "$ACTION" = remove runs the following while loop:
while read device mountpoint fstype remainder; do
if test "$DEVNAME" = "$device"; then
...
umount -l "$mountpoint"
...
run-parts /etc/usbmount/umount.d || :
...
fi
done < /proc/mounts
Unfortunately for this loop, udev (the only possible culprint) has
already removed the node /dev/sda1 by the time this is called. And as
side effect of doing that has also umount-ed it. So /proc/mounts does
not anymore contain an line for the device, and the "if" inside the
while fails to trigger. Thus the run-parts never gets executed.
Note: The umount -l also does not get executed, but this does not
show, als udev has already done so. This is also most likely the reason
why this bug was overlooked in testing. In the default install it only
shows itsself in form of littering of /var/run/usbmount. When users add
their own scripts to /etc/usbmount/umount.d the bug then manifests
itsself.
Suggested Fix:
While running the "$ACTION" = add case locally store all the
information needed for "$ACTION" = remove so there is no dependency
on /proc/mounts.
Note: You can not use /etc/mtab either, as that has also allready been
cleaned up by the umount done by udev.
Note 2: This is more a workaround than a true fix. Properly the udev
people should not remove /dev/sda1 before all /etc/dev.d/ scripts have
run. But the udev team has already eliminated /etc/dev.d/ alltogether
in newer versions (they are up at 084), so they will not care about an
bug report on this.
--
Neil Franklin, System Programmer, Dept of Physics
Swiss Federal Institute of Technology (ETH), Zuerich, HPR E86.1
http://www.phys.ethz.ch/~franklin/, [EMAIL PROTECTED]
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
