This is a problem with the dhcp daemon. If the disk is full, the dhcpd daemon dies because it can't write to the lease file. My solution is to add dhcpd daemon to a reserved blocks group so that it will be able to access the last 5%(default) of the disk and be able to write the lease file insted of dieing. This would mean that the reserved blocks group would have to be changed from root to something else either by the ebild or notifying the user to add it to the special reservedblocks group with tune2fs -g reservedblocks /dev/hdX.
Uh, bad idea really... clean up your drive.
This is a bad idea because it would open you up to a denial of service attack from a sort of trusted user. If they are trusted enough to request a dhcp lease they still shouldn't be allowed to bring the system down completely filling the disk. Basically, if the drive doesn't have the room to register the access then it should deny the access else there is an un-logged user on the system/network. If you allow it to completely fill the drive (by writing into the area reserved for root) then it could shut off even more important services. If it really is a problem for you then create a dhcp user and use quotas to keep someone from consuming more than their fair share of the disk.
*** Bug 144852 has been marked as a duplicate of this bug. ***