Alexey Mahotkin wrote:
It almost cleanly applies to cvs-1.10.8 and, as the matter of fact, the new
release of cvs-nserver will be against 1.10.8. The most significant
modification of original code is the removal of about 600 lines from
server.c, yet it is still way, way too long.
Ahh... server.c is the most heavily modified part of cvsnt (Essentially I had to
rewrite
half of it to use threads rather than forking). Much fun ahead, methinks...
There is an obvious task to improve server.c by splitting kerberos- and
GSSAPI-related code from it thus creating cvs-kserver and cvs-gserver.
There is probably also need to create cvs-sslserver (I have not
investigated yet whether we could get along with ssl-tunnel'ed server (we
surely can not get along with ssl-tunneled client as it almost has nothing
to tunnel)).
For NT you would also need cvs-ntserver. It might be worth investigating whether
cvs-kserver could be ported to NT too (although the MS documentation on this is worse
than useless).
It seems to me that checkpassword scheme is sub-perfect for NT though I
could be wrong. I've tried to research security aspects of NT but has not
reached considerable results. And after I learned about your project and
changed job recently hoping not to see MS in a lifetime no more (though it
seems like I will have to anyway) I completely relaxed and thought that I'd
be better off with CVS under UNIX. Though I will be glad if nserver will
influence the development of NT-CVS or vice versa.
Under NT you can't do setuid, and you can't check against a pre-encrypted system
password. The only
way to validate a password is to attempt a non-interactive login (after which you can
change you UID
to it). Of course this means you need the original plain-text password to work, and
this has security
implications. There isn't a way around this as far as I can see.
Tony
--
#define QUESTION ((bb) || !(bb)) - Shakespeare
[EMAIL PROTECTED]
