openssh unix dev - Jul 2001 - RFD: uid of privileged user not fixed to 0

Hi,

I don't know if other systems are concerned as well but it's at least
a problem for Cygwin that the uid of the privileged user is fixed to
the constant 0 in the OpenSSH sources.

A month ago I ported Paul Vixie's cron to Cygwin and he's using
at least a define for that:

	#define ROOT_UID 0

which simplified the porting vastly.

May I propose to change the usage of uid 0 in the sources to become
substituted by a define as above which would allow to configure
that for different systems which doesn't neccessarily have the uid
0 and a root user at all?

Instead of constant (and badly portable) comparisons like

	if (st.st_uid == 0)

we could use something like

	if (is_root(st.st_uid))

which would allow macros like the following on Unices:

	#define ROOT_UID 0
	#define is_root(x)	((x) == ROOT_UID)

and other definitions on other systems:

	#define ROOT_UID 18	/* "SYSTEM" account on NT/W2K */

and even functions:

	#define is_root(x)	wierd_system_is_root(x)
	extern int wierd_system_is_root(uid_t);

Is there any problem to do so? Since that change isn't messing
anything up (IMO) it should be no problem to apply that already
to the OpenBSD base sources of OpenSSH.

Any comments (besides flaming on Windows)?

Corinna

-- 
Corinna Vinschen
Cygwin Developer
Red Hat, Inc.
mailto:vinschen at redhat.com

I support the idea in general.
>I don't know if other systems are concerned as well but it's at
least
>a problem for Cygwin that the uid of the privileged user is fixed to
>the constant 0 in the OpenSSH sources.
Other systems I can think of are systems that use fine grained privileges
rather than uid 0 to determine access - but those systems are likely to
have their own additional source modifications as well. (I'm thinking of
things like Trusted {Solaris, BSD}, DEC MLS).
>we could use something like
>
>	if (is_root(st.st_uid))
What about using suser() which is what is used in many kernels to
do the uid 0 check ?

suser() on Solaris only exists in kernel space (and is actualy a function
since it has auditing and accounting side effects) but the macro you would
use in userland is just as you described for is_root().

--
Darren J Moffat