Hi Bill, > On Tue, Jul 23, 2002 at 10:13:37PM +0200, Michael Hoennig wrote: > > That IS the reason, and I found the cause: There was a nohup job > > writing to nohup.out, which was deleted every 30 minutes by a cronjob > > instead of truncated. These dead nohup.outs were not reported by lsof, > > and of cause not linked in the file system anymore, but yet using up > > space. Isn't the last point a bug? > > Of course not; you have a file descriptor to which you can (in the > general case) read/write/seek arbitrarily. It needs to be stored > *somewhere* and that is in the filesystem. This behaviour has been used > for implementing temporary files that are automatically cleaned up (see > tmpfile(3)) upon file close / program exit for a *long* time. ok, then at least lsof should report such files. Actually I forgot something: Even after all processes holding this file handle, the storage was never freed, only after umount/mount! I have not checked it yet, but if this storage is not even counted in quota, which I guess is the case, it is a possible local DoS. Any user, even with a reasonable quota, could flood the device. Michael -- Hostsharing eG / Boytinstr. 10 / D-22143 Hamburg phone+fax:+49/700/HOSTSHAR(ing) (= +49/700/46787427) http://www.hostsharing.net: host, where YOU make the difference
