Matus fantomas Uhlar wrote:
> -> on Tue, Jan 21, 2003 at 06:45:30PM -0600, Hanasaki JiJi wrote about Re: autofs vs 
>amd:  Is there a preference?:
> -> > anyone have a config for amd to automount home dirs?  off redundant 
> -> > hostnames?
> -> > 
> -> > in amd, when fileserver:/exports/[export1] is automounted, the 
> -> > files/dirs in [export1] are accessable but not viewable in a "ls" Any 
> -> > way to change this?

Hmm... I see that the directory containing [export1] is not
browsable, by design.  But files in [export1] are browsable.  Some
concrete examples would clear the confusion up.  But until then I will
say "works for me".

> -> No. This is just how the automounter works. Can you imagine what kind
> -> of a "mount-storm" you would have if someone just used an ls, or tab
> -> completion in the parent directory (in your case /mnt/dir1)? FYI, the
> -> O'Reilly book "Managing NFS and NIS" explains this quite well.

Having tested that configuration it is rather interesting to have 1500
NFS mount points.  Things don't work as well there.  (This was on an
hpux host.)  I can't recommend it but have not tried it on the Linux
kernel.

> OTOH, it is currently not possible to mount a FS without explicitly changing
> to nonexistent directory. You can workaround this with symlink that points
> to the destination, but in that case never call stat() on that link.

Huh?  If you access the trigger point of the mount it will be
mounted.  You can mount it by doing any of 'test -f file', 'ls' or
anything.  We must be talking about different things to be this out of
sync.

> imho, the directory should exist (if it's possible) and stat() should not
> cause mount, open and chdir should do. iirc it is in the TODO list for
> automounter.

It is not possible.  When you stat() a file you will need to get the
inode information which is stored on the NFS server not on the client
automounter.  The automounter must mount the underlying disk in order
to get that information.  (This is one cause of "mount storms" in the
above mentioned not-recommended configurations.  Running 'ls -l' on a
symlink farm will stat() each and mount all.)

Bob

Attachment: msg26212/pgp00000.pgp
Description: PGP signature

Reply via email to