[Gluster-devel] posix-locks problem

[Anand Avati]

2007/10/22, Kevan Benson <kbenson at a-1networks.com>:
>
> Vincent Régnard wrote:
> > Using this latter script reveals that locking is working fine :)). Using
> > my program properly also works. I was actually removing the lock file
> > when it was locked by another instance ! So my programm was misbehaving
> > after that. But the question is "why is it possible to remove a file
> > when it is locked?". I can imagine the answer is that system call is not
> > the same. Should'nt glusterfs translator forbid or delay unlink() on a
> > file when it knows a file is locked ?


fcntl or flock based locks are on file content and not on the namespace.
unlink() is a namespace operation and no locks are supported by any standard
(yet) on the filesystem namespace. This is because even when the file is
unlinked (which is actually 'removing a directory entry'), the "actual file"
may still be reachable via other hardlinks or open file descriptors. So it
does not make sense to forbit unlink()s.

avati


-- 
It always takes longer than you expect, even when you take into account
Hofstadter's Law.

-- Hofstadter's Law