there is no way to enforcing bsd group semantics on nfs mounts

[ghost]

Metadata

• Number: 7731
• Originator: Luke Mewburn <lukem@netbsd.org>
• Arrived: Mon Jun 07 17:50:00 +0000 1999
• Last modified:
• Environment: System: NetBSD karybdis.cs.rmit.edu.au 1.4C NetBSD 1.4C (LUKEM) make depend fails to run mkdep sometimes. #139: Tue Jun 1 11:36:43 EST 1999 lukem@karybdis.cs.rmit.edu.au:/echidna/netbsd/src/sys/arch/i386/compile/LUKEM i386

Description

    I recently migrated from a local copy of /usr/src to one NFS mounted
    off one of my NetApp filers.

    I noticed that new files on the NFS mount are being created with
    my primary group (SYSV behaviour), instead of the group of the
    parent directory (BSD behaviour).

    The workaround is to use what I used on my Solaris clients for
    a long time; set the setgid bit on the parent directory. I don't
    need this anymore on Solaris since I discovered the `grpid' mount
    option which enforces BSD semantics.

    The setgid behaviour wasn't documented anywhere in NetBSD as far
    as I can tell (chmod(1), chmod(2), etc).

    I find it ironic that a SYSV derived system can enforce BSD
    semantics in this situation, but a BSD derived system requires
    a SYSV hack to function correctly. ;-)

    I don't want to setgid all my directories again; setgid on a
    directory has a different lossage mode under Solaris (if the
    umask is set to mask out group execute, new directories will
    be made with g+l (mandatory locking) instead of g+s (setgid),
    as the 2000 bit is inherited by the directory in all situations.
    This is lame, but not our bug :)

How to repeat

    NFS mount a directory off a NetApp fileserver (and maybe others?)
    Create some files. Notice that they have your primary group as
    their group unless the parent directory has the setgid bit set.

Fix

    I suggest that we either:
        * fix our NFS client so that it always has BSD semantics
    or
        * provide a `grpid' mount option for mount_nfs (yuck!)