[Fwd: Re: nfs-utils]

DJ Lucas dj at linuxfromscratch.org
Wed Apr 20 22:06:45 PDT 2005


Bruce Dubbs wrote:
> 
> Nathan,
> This seems like pretty non-standard use to me.  It took me a while to
> figure it out.  I'm sure its something that would suprise others too. Is
> there a reason that we need to avoid mentioning specific filesystems in
> the scripts?
> 
>   Here is what I see as a summary of the alternatives:
>   1.  Train all users that use remote filesystems to put a _netdev
> option in fstab for all network mounted partitions.
> 
>   2.  Change the mountfs script to say:
>       mount -a -t nonfs,ncp,smb -O no_netdev
> 
>       and change the netfs script basically to say:
>       mount -a -t nfs
>       mount -a -t ncp
>       mount -a -t smb
>       mount -a -O _netdev
> 
>   Before we make changes, I would like to get other opinions.
> 
>   -- Bruce

Bruce, the goal in mind at the time was to avoid changing the script as
more network filesystems were added (coda, cifs, and afs (andrew???)
IIRC).  This is really not a big issue as I don't see a lot of new
network types floating around out there in the world, but simply adding
_netdev to the fstab automatically lets the script know that it's a
newtork filesystem and should be acted upon.

Things to keep in mind for rewriting the script:  samba is broken anyway
WRT _netdev and doesn't keep the option in the the mount list so it was
specifically worked around for the  umount case.  Use of 'awk' or 'cut'
is a bad idea unless used only for creating a list of filesystems, but
the complete list must be generated before the umount actually happens
(/usr).  Last, we should properly kill running processes before the
umount happens ('fuser -km', but this has caused locking problems with
nfs in the past IIRC).  Jim Gifford had a suggested replacement in his
svn repo some time ago, if it's still availible, it should be examined.

-- DJ Lucas



More information about the blfs-dev mailing list