I?m seeing the following situation occur with 2.4.23 up through 2.4.27 and
autofs 4.1.3 using /net access with a generic /etc/auto.net. I tried
'nonstrict' and 'nosymlink' in /etc/auto.net but that resulted in the
mounts not working.
NFS Server foo has virtual nested exports:
/ on /dev/sda1
/bar on /dev/sdb1 (for example)
cd /net/foo/bar mounts both / and /bar which it finds from ?showmount ?e?
df shows foo:/ and foo:/bar mounted
/proc/mounts shows both, also.
I?m guessing what?s happening next is autofs4 tries to umount foo:/ before
foo:/bar which yields a busy condition for foo:/, even though it?s not
because foo:/bar is mounted above foo:/ on the client.
/proc/mounts starts filling up with foo:/bar entries at this point and
fills up the mount table.
If I bring down the automount process, manually go through and umount all
of the foo:/bar entries in /proc/mounts, clean up anything in /etc/mtab and
then restart the automounter, everything seems to be OK. Obviously, this
is not acceptable.
Any ideas?
On Mon, 13 Sep 2004, Mitch Sako wrote:
> I?m seeing the following situation occur with 2.4.23 up through 2.4.27 and
> autofs 4.1.3 using /net access with a generic /etc/auto.net. I tried
> 'nonstrict' and 'nosymlink' in /etc/auto.net but that resulted in the
> mounts not working.
>
> NFS Server foo has virtual nested exports:
> / on /dev/sda1
> /bar on /dev/sdb1 (for example)
>
> cd /net/foo/bar mounts both / and /bar which it finds from ?showmount ?e?
>
> df shows foo:/ and foo:/bar mounted
>
> /proc/mounts shows both, also.
>
> I?m guessing what?s happening next is autofs4 tries to umount foo:/ before
> foo:/bar which yields a busy condition for foo:/, even though it?s not
> because foo:/bar is mounted above foo:/ on the client.
>
> /proc/mounts starts filling up with foo:/bar entries at this point and
> fills up the mount table.
>
> If I bring down the automount process, manually go through and umount all
> of the foo:/bar entries in /proc/mounts, clean up anything in /etc/mtab and
> then restart the automounter, everything seems to be OK. Obviously, this
> is not acceptable.
>
> Any ideas?
Are you using a kernel with the latest autofs4 patch?
btw the place for these questions is the autofs mailing list.
Ian
On Mon, 13 Sep 2004, Mitch Sako wrote:
> I?m seeing the following situation occur with 2.4.23 up through 2.4.27 and
> autofs 4.1.3 using /net access with a generic /etc/auto.net. I tried
> 'nonstrict' and 'nosymlink' in /etc/auto.net but that resulted in the
> mounts not working.
>
> NFS Server foo has virtual nested exports:
> / on /dev/sda1
> /bar on /dev/sdb1 (for example)
>
> cd /net/foo/bar mounts both / and /bar which it finds from ?showmount ?e?
>
> df shows foo:/ and foo:/bar mounted
>
> /proc/mounts shows both, also.
>
> I?m guessing what?s happening next is autofs4 tries to umount foo:/ before
> foo:/bar which yields a busy condition for foo:/, even though it?s not
> because foo:/bar is mounted above foo:/ on the client.
autofs should handle this OK.
>
> /proc/mounts starts filling up with foo:/bar entries at this point and
> fills up the mount table.
This should be easy to duplicate. Is there anything else I should know
about this setup?
Ian