What on earth is mounted?
Daniel J Walsh
dwalsh at redhat.com
Tue May 31 14:47:45 UTC 2011
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 05/26/2011 11:43 AM, Michael Hennebry wrote:
> On Wed, 25 May 2011, Michael Cronenworth wrote:
>
>> Harald Hoyer wrote:
>>> Most of them are from the sandbox init script .... NOT from systemd!!
>>
>> Yes, this was not documented in the release notes. It is part of
>> SELinux/pam_namespace usage.
>
> I've been looking, but I haven't found it.
> If I do find it, will it answer the question?
>
> On Wed, 25 May 2011, Tom Horsley wrote:
>
>> I was checking to see if I had everything mounted that I wanted
>> in a newly genned f15 system, and I see this insanity in
>> the output from running "mount":
>>
>> /dev/sda2 on / type ext3
>> (rw,relatime,errors=continue,user_xattr,acl,barrier=$
>> /dev/sda2 on /tmp type ext3
>> (rw,relatime,errors=continue,user_xattr,acl,barri$
>> /dev/sda2 on /var/tmp type ext3
>> (rw,relatime,errors=continue,user_xattr,acl,b$
>> /dev/sda2 on /home type ext3
>> (rw,relatime,errors=continue,user_xattr,acl,barr$
>>
>> Where on earth do those /tmp /var/tmp and /home entries come from?
>> They certainly aren't all mounted on top of the same filesystem root.
>> There are no entries for them in /etc/fstab. What is going on?
>
> Whatever is doing it, what is being done?
> Is the same partition really on four different mount points?
> To me, the above looks rather scary,
> rather like my first encounter with LVM.
> It was in the middle of an install.
>
> To me, the most important thing that should be in release
> notes is what will break from an extant release.
> The next is what will look different enough to be scary.
> I'm not sure which this falls under.
>
This has been like this for many releases. All the way back to RHEL5,
has this.
There is an open bug on pam_namespace that will hopefully eliminate the
need for this.
When you use namespaces on mount tables you need to separate out file
systems, so you can say something like I want these file systems shared
with the entire system, while these other file systems private the each
namespace. pam_namespace sets up a private namespace on $HOME and /tmp.
In order to do this is needs /tmp and $HOME to be on their own file
system. Therefore the init script bind mounts /tmp on /tmp and /home
on /home causing the problem you are seeing. It also executes the
appropriate commands to make "/" shared and /tmp and /home private.
If you do not use pam_namespace or sandbox you can disable the init
script and on the next boot the problem will go away.
sandbox tool now is smart about setting up these private/bind mounts
when you run the sandbox, rather then making them the system defaults.
I wrote a patch for this for pam_namespace, when the patch is applied
to pam_namespace, I will ask systemd to setup "/" as shared and remove
the init script altogether.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
iEYEARECAAYFAk3k/5EACgkQrlYvE4MpobNHaQCgt9yij/MY7/E5WanpB/UD6EM7
/zcAn3Fla09leAQXW8JFAcD9iGgJ+4uA
=w7Vs
-----END PGP SIGNATURE-----
More information about the users
mailing list