[ros-dev] dealing with insecure-by-inattention Windows software

KJKHyperion hackbunny at reactos.com
Wed Dec 14 03:20:25 CET 2005

Wesley Parish wrote:

>"insecure-by-inattention" - by that I mean software that must run as super-user or otherwise (otherstupidly) it won't run at all.
read all you can find on the new security features in Windows Vista. The 
security model has been enhanced with various forms of Mandatory Access 
Control - among the many of them, the fact that all users (except 
System, i.e. services) are entirely powerless by default, even 
administrators, and a reauthentication is required every time privileges 
are needed (the user experience is pretty much identical to MacOSX) - 
and the application compatibility layer has been enhanced to the point 
it now fully implements a virtual filesystem view and a virtual registry 

As for per-application sandboxing, it's very possible even today (with 
restricted tokens) and I use it daily. It has a couple of issues, one of 
which strictly technical and the others concerning usability. The 
technical issue is that sandboxes are barred from any form of 
client-side SSPI authentication - this means no accessing CIFS shares, 
no integrated authentication to an ISA proxy server, no remote 
administration through a snap-in, etc. (all attempts to create outbound 
credentials fail with a spurious "out of memory" error). It's hard to be 
worked around and presumably it's there for a reason other than "it was 
too hard to do".

The usability issues are many, many, many. Too many to be contained in a 
mailing list posting. I think it'd help to describe how an end user 
should interact with it, and then code whatever hacks are required for 
it, or rethink the interaction if the hacks prove to be too ugly.

The problem with your approach? too much maintenance. Even for a server. 
Setup packages for Windows tend to be a mistery, a surprise after 
another, a gift that keeps giving, can you really tell how to categorize 
their files and registry keys? and even then, are you sure that'd be a 
good idea?

More information about the Ros-dev mailing list