On Thu, 2005-01-20 at 19:56 +1100, Nick Urbanik wrote:
This raises a can of worms when maintaining the program, and the
question arises as to which is the "real one".
Well...no, since you still have the same source code and build process,
etc. This solution is a lot like what pre-SELinux chroot scripts did
for bind, etc.
I'm likely to forget
to update one or the other.
I'd imagine that your Makefile or whatever would install the two copies
explicitly. Or you could do it in the RPM build process.
"Which one do I enter into version
control?" is a question I would ask myself often.
You enter binaries into version control?
Where are SELinux attributes stored? In the inode?
They are tightly coupled to the inode, yes. Just like Unix permissions
are.
If not, can hard
links be given different attributes?
No; hard links are just additional names for the same object. SELinux
protects the actual object, not names or references to objects.
> The other solution is to define a new type, and grant both
domains in
> question access to it. This is a lot more complex; now you have to
> consider potential information flow between the two domains which were
> (presumably) separate before.
Well, that may be more managable in the long term. Can you suggest a
(relatively) simple way of doing that?
You'd have to explain more about your setup. Are you just trying to run
the CGI script as an ordinary user from unconfined_t?