Re: [PATCH 0/3] vfs: plug some holes involving LAST_BIND symlinks and file bind mounts (try #5)

[ebiederm@xmission.com (Eric W. Biederman)]

Pavel Machek <pavel@ucw.cz> writes:

> On Tue 2009-11-24 13:59:06, Miklos Szeredi wrote:
>> On Tue, 24 Nov 2009, Pavel Machek wrote:
>> > I believe that current semantics is ugly enough that 'documenting' it
>> > is not enough... and people want to port from other systems, too, not
>> > expecting nasty surprises like this...
>> 
>> This hasn't been a problem for the last 12 years, and still we don't
>> see script kiddies exploiting this hole and sysadmins hurrying to
>> secure their system, even though it has been public for quite a while.
>> 
>> Why?
>
> Because condition when it hits are quite unusual?

So unusual perhaps that this is not a problem?

>> The reason might be, that there *is no* violation of security.
>
> Well, security people disagree with you.

Other security people disagree with you.

>> See this: the surprise isn't that an inode can be reached from
>> multiple paths, that has been possible with hard links for as long as
>> unix lived.  The suprise is that the inode can be reached through
>> proc.  So this "hole" that has been opened about 12 years ago in linux
>> is quite well known.  Only this particular aspect of it isn't well
>> known, but that doesn't mean it's not right, does it?
>
> It does. Bypassing checks on read-only file descriptors is design
> misfeature, and users are clearly unaware. (See bugtraq). Being "old"
> does not mean it is right.

Being "old" does mean that changing it is a regression if any valid
application depends on this feature.

Eric