Re: [LTP/VFS] fcntl SETLEASE fails on ramfs/tmpfs

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Apr 29, 2008 at 06:01:47PM -0400, Mike Frysinger wrote:
> On Tue, Apr 29, 2008 at 5:42 PM, J. Bruce Fields <bfields@fieldses.org> wrote:
> >  The one thing I suspect is *not* a really serious problem here is the
> >  reported LTP failure, since probably the only user of this is Samba,
> >  which probably doesn't do a lot of tmpfs exports, and in any case it can
> >  probably soldier on (if with degraded performance--how badly I don't
> >  know) without getting the write lease it wants.
> 
> i'm not sure i follow.  the reported problem is that file locking does
> not work on tmpfs/ramfs storage.  a not terribly uncommon scenario is
> to use tmpfs on /tmp (or similar location) and have file locking not
> work at all.  so programs that use file locking or scripts that
> leverage the flock utility from the util-linux package break.

There are three different mechanisms that might be called "file locks":

	- fcntl() locks, aka "posix locks", "byte-range locks":
	  documented in the "Advisory locking" section of fcntl(2).
	- flock() locks: documented in flock(2) This is what the shell
	  utility from util-linux uses.
	- leases, documented in the "Leases" section of fcntl(2).

For the former two, I agree with you, applications actually depend on
them.

The bug report, however, is for leases, which are much less widely used.
(The only users I know of are Samba and, to a lesser extent, NFSv4
(which doesn't currently use write leases due to all these problems).)

--b.