What's wrong with tmpfs?

What's wrong with tmpfs?

[Rob Landley]

Under User Mode Linux in 2.6.14, would someone please explain to me why this 
test:

static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
{
        int err;

        if (mnt->mnt_sb->s_flags & MS_NOUSER)
                return -EINVAL;

Is triggering when I try to mount tmpfs?  Is this happening for anybody else?  
Shouldn't I be getting a fresh superblock or something?  (Is this just a User 
Mode Linux issue?  Haven't got a spare box set up to boot it on real hardware 
just yet...)

If somebody needs a reproduction sequence, I'm happy to oblige.  In theory 
"mount -t tmpfs /mnt /mnt" should do it, but if it was _that_ simple it 
wouldn't have shipped...

Rob

Re: What's wrong with tmpfs?

[Jon Masters]

On 10/30/05, Rob Landley <rob@landley.net> wrote:

> If somebody needs a reproduction sequence, I'm happy to oblige.  In theory
> "mount -t tmpfs /mnt /mnt" should do it, but if it was _that_ simple it
> wouldn't have shipped...

I don't see this behaviour on a regular desktop box running 2.6.14.
Guess it's UML specific.

Jon.

Re: What's wrong with tmpfs?

[Jeff Dike]

On Sun, Oct 30, 2005 at 12:53:00PM +0000, Jon Masters wrote:
> On 10/30/05, Rob Landley <rob@landley.net> wrote:
> 
> > If somebody needs a reproduction sequence, I'm happy to oblige.  In theory
> > "mount -t tmpfs /mnt /mnt" should do it, but if it was _that_ simple it
> > wouldn't have shipped...
> 
> I don't see this behaviour on a regular desktop box running 2.6.14.
> Guess it's UML specific.

Sorry, but wrong.

IIRC, this triggers when you don't have CONFIG_TMPFS enabled.  If you don't,
you still get it, but you get a version that's only usable in-kernel.

				Jeff

Re: What's wrong with tmpfs?

[Rob Landley]

On Sunday 30 October 2005 09:15, Jeff Dike wrote:
> On Sun, Oct 30, 2005 at 12:53:00PM +0000, Jon Masters wrote:
> > On 10/30/05, Rob Landley <rob@landley.net> wrote:
> > > If somebody needs a reproduction sequence, I'm happy to oblige.  In
> > > theory "mount -t tmpfs /mnt /mnt" should do it, but if it was _that_
> > > simple it wouldn't have shipped...
> >
> > I don't see this behaviour on a regular desktop box running 2.6.14.
> > Guess it's UML specific.
>
> Sorry, but wrong.
>
> IIRC, this triggers when you don't have CONFIG_TMPFS enabled.  If you
> don't, you still get it, but you get a version that's only usable
> in-kernel.

Yup.  Not CONFIG_TEMPFS.  My bad.

Rob

Re: What's wrong with tmpfs?

[Matt Mackall]

On Sun, Oct 30, 2005 at 10:15:06AM -0500, Jeff Dike wrote:
> On Sun, Oct 30, 2005 at 12:53:00PM +0000, Jon Masters wrote:
> > On 10/30/05, Rob Landley <rob@landley.net> wrote:
> > 
> > > If somebody needs a reproduction sequence, I'm happy to oblige.  In theory
> > > "mount -t tmpfs /mnt /mnt" should do it, but if it was _that_ simple it
> > > wouldn't have shipped...
> > 
> > I don't see this behaviour on a regular desktop box running 2.6.14.
> > Guess it's UML specific.
> 
> Sorry, but wrong.
> 
> IIRC, this triggers when you don't have CONFIG_TMPFS enabled.  If you don't,
> you still get it, but you get a version that's only usable in-kernel.

That sounds like a regression. Turning off CONFIG_TMPFS replaces tmpfs
with an aliased ramfs. It should be perfectly usable everywhere that
tmpfs is, with the exception that it's not swap-backed and doesn't
have an size limiting.

-- 
Mathematics is the supreme nostalgia of our time.