[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <45AF6109.3030607@aim.com>
Date: Thu, 18 Jan 2007 06:59:05 -0500
From: Rage Coder <ragecoder@....com>
To: 3APA3A <3APA3A@...URITY.NNOV.RU>
Cc: bugtraq@...urityfocus.com
Subject: Re: Windows logoff bug possible security vulnerability and exploit.
The problem only occurs at times. To reproduce the problem, I just use
the computer normally, and at each logon check the event viewer and
running processes to see if a profile unload failed. I don't have any
special terminal software or other logon software installed.
I find that if I wait for a little bit after logging off before logging
on again, no running programs from the previous logon are present, but
if I log on just after logging off, they will be if the profile unload
fails. That still shouldn't be the case. My brother frequently goes on
his account right after I go off; there shouldn't be a time limit to
wait in order to prevent this.
I noticed an interesting thing about XP and fast user switching which
would likely stop this problem. When logging on, the first logged on
user is given session ID 0, as shown in task manager, but if I 'switch'
to another user, the user is given a different session ID. It seems
that no two users are given the same session ID when using fast user
switching. But when logging off all users and then back on, it is back
to session 0. And if I just log on as a user, log off, and then on as
another user without using the 'switch user', they both are session ID 0.
The same thing happens when using classic logon and on 2003. All logons
are given session ID 0. I did some reading in the platform SDK and some
sites about stuff, and it seems that these sessions literally create an
isolation. Messages sent from a process in one session ID are not
visible to processes in another, windows created only appear on the
desktop associated with that session of the process that created the
window, etc.
Ideally, running classic logon always as session 0 'should' work because
ideally when logging of, the processes ran 'should' close, so the next
user to log on would have nothing to access. But this does not appear
to be the case at all times.
A few moments ago I logged in as administrator to do some minor changes,
and I ran EPIM to take some notes of things. When I logged of and back
on as a regular using, 'explorer.exe', 'essentialpim.exe',
'seamonkey.exe' were still running as Administrator, event viewer showed
the usual UserEnv messages, and EPIM appeared on the system tray. My
guess is something like this happens:
Logon Administrator : Session ID 0
Run EssentialPIM : Session ID 0
Do some stuff
Logoff Administrator : Profile unload fails, a few programs continue running
Logon Normal User : Session ID 0
Explorer runs, and at startup broadcasts 'TaskbarCreated' message
All processes in session 0 get this message, EPIM adds system tray icon
like it is supposed to
If each logon, even in classic mode, is given a separate session ID as
is done in fast user switching, this would not happen, even if the
profile unload fails and the programs continue to run waiting for the
profile to unload:
Logon Administrator : Session ID 0
Run EssentialPIM : Session ID 0
Do some stuff
Logoff Administrator : Profile unload fails, a few programs continue running
Logon Normal User : Session ID 1
Explorer runs, and at startup broadcasts 'TaskbarCreated' message
All processes in session 1 get this message
Programs that may continue to run in session 0 are isolated
If I log on as administrator again, it would be ok to reuse session 0,
but for a given boot, no two users should be assigned the same logon
session ID. I.E. if I log on as Normal User again, it would be session
1, etc.
This would not prevent a profile from failing to unload, and would not
prevent the processes from continuing to run, but it will prevent a user
from a later logon from accessing the processes in the current logon.
3APA3A@...URITY.NNOV.RU wrote:
> Dear Rage Coder,
>
> I've seen unloaded profiles for many times, but I never saw application
> still running after logoff. Profile itself doesn't create security
> vulnerability, since it can not be accessed by another user.
>
> What do you use to reproduce this vulnerability?
>
> Are you sure you do not use some different software which affects
> logon/logoff process, e.g. 3rd party terminal software or some security
> enhancement?
>
>
Powered by blists - more mailing lists