Re: /proc/pid/fd && anon_inode_fops

[Oleg Nesterov <oleg@redhat.com>]

Hi Willy,

On 08/24, Willy Tarreau wrote:
>
> On Sat, Aug 24, 2013 at 08:29:39PM +0200, Oleg Nesterov wrote:
> >
> > On 08/22, Willy Tarreau wrote:
> > >
> > > It's not only that, it also supports sockets and pipes that you can access
> > > via /proc/pid/fd and not via a real symlink which would try to open eg
> > > "pipe:[23456]" instead of the real file.
> >
> > But sock_no_open() disallows this, and for good reason I guess.
>
> Hmmm not exactly, it works for a pipe but not for a socket.

But this is what I meant, sorry for confusion.

Let me try to explain. Just in case, this has nothing to do with security
and I do not see any problem, still I think there is something wrong (but
harmless).

Suppose that you are trying to open(/proc/pid/$pipe-or-socket-fd).
nd_jump_link() sets nd->inode correctly, then dentry_open() does the
rest. Everything is correct at this stage, the new file gets the correct
f_inode/f_op.

However, unlike fifo_open(), socket_file_ops->open() can not actually
create the file/sock connection, so sock_no_open() just fails and
nothing bad happens.

But if you open an anon_inodefs file via proc, you get the "bogus" file.
There is a single anon_inode_inode, its ->i_fop points to the empty
anon_inode_fops, this has nothing to do with ->f_op of the actual file
you tried to open.

Nothing bad happens, still I think this is simply wrong and misleading,
and thus I think it would be better to disallow this via anon_open().

Oleg.