On Thu, 24 Aug 2006, Matt Bernstein wrote:
> Hi,
>
> We like to preconfigure an number of our (1000ish) users' applications.
> Typically, this might necessitate hard-coding their home directories in
> dot-files, so we specify their home directory in the passwd map as
> /homes/$LOGNAME, and make this an (LDAP) automount map which might point
> to ":/server/{servername}/{volumename}/$LOGNAME" (via a bind mount), where
> /server/{servername}/{volumename} would be the "real" NFS mount.
>
> This allows us to move the physical location of any user's home directory
> without upsetting their application settings. (Some applications seem to
> prefer using full paths.)
>
> Our scheme all worked fine in autofs 3.
>
> In autofs 4, it worked but with a 10 or 15s delay which we got around via
> the following patch:
>
> --- modules/mount_bind.c.orig 2004-10-29 14:46:11.000000000 +0100
> +++ modules/mount_bind.c 2004-11-01 12:49:54.000000000 +0000
> @@ -126,7 +126,10 @@
> }
>
> debug(MODPREFIX "calling mount --bind %s %s", what,
> fullpath);
> -
> +/* 'what' may itself be on an automounted file system so we need to access
> it */
> +/* to ensure it gets mounted before calling wait_for_lock() inorder to avoid
> a*/
> +/* long delay */
> + (void)access(what,F_OK);
That can lead to corruption of /etc/mtab.
That's why the lock was introduced.
In v5 the locking can be disabled as most people have updated mount(8)
these days.
> wait_for_lock();
> err = spawnl(LOG_NOTICE, MOUNTED_LOCK,
> PATH_MOUNT, PATH_MOUNT, "--bind",
>
>
> In autofs 5, it doesn't work at all, claiming no hostname was supplied. Is
> there any way we can achieve what we want? I can't find any documentation.
Someone else has reported that bind mounts aren't working with v5.
It would be good to resolve this but I need more info.
A debug log would be best.
Ian
_______________________________________________
autofs mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/autofs
