flock() over NFS

[Brian Long <brilong cisco com>]

Has anyone else experienced difficulty with applications that expect
flock() to work over NFS?  We have various build environments built
around NFS and ClearCase.  Many makefiles use a flock program which
calls flock() and works over NFS on all distributions except RHEL 3
(works on Debian Sarge, FC3, RHEL 4 beta, RH 7.1, 7.3, Suse 9.2).

The developers are not trying to lock a single file for multiple
machines over NFS.  They're just trying to lock the file knowing the
file will only be accessed by the one machine, but it happens to be over
NFS.

The flock() man pages says flock doesn't work over NFS (even RH 7.1 man
page says this, but it returns success anyways).  We understand it
shouldn't work with multiple clients trying to lock or check status for
the lock on a single file, but what about a single client running a make
-j8 over NFS?  Why is flock() only broken in RHEL 3?

When we opened a case with IBM GS which back-lined to Red Hat Level 3,
they said it was working as documented and they wouldn't fix it.  I'm
trying to understand why it's only affecting us on the RHEL 3 platform.

FYI, if we mount the NFS filesystem with -nolock, this gets around the
behaviour, but it also opens up the posibility of problems for programs
using fcntl().  Any ideas on how to resolve this would be appreciated!

/Brian/
-- 
       Brian Long                      |         |           |
       IT Data Center Systems          |       .|||.       .|||.
       Cisco Linux Developer           |   ..:|||||||:...:|||||||:..
       Phone: (919) 392-7363           |   C i s c o   S y s t e m s