Home page logo
/

bugtraq logo Bugtraq mailing list archives

Re: [FD] Beginner's error: import function of Windows Mail executes rogue program C:\Program.exe with credentials of other account
From: "Stefan Kanthak" <stefan.kanthak () nexgo de>
Date: Wed, 30 Jul 2014 17:21:07 +0200

"Joe Souza" <Joe.Souza () NetMotionWireless com> wrote:

It is at the very least ignorant to call Microsoft's CreateProcess
behavior "braindead".

What else is it then?

If anything it shows your complete lack of understanding of the issue.

Really? Let's see how good your understanding of the Win32 API and its
compatibility to the Win16 API is.

The behavior of the API is not arbitrary; Microsoft has always had to
walk the line between security and application compatibility.

And every so often they made the wrong decision!

If Microsoft were to change behavior of CreateProcess as you suggest,
then they would break untold apps that have been written so as to call
the API without quoting the target executable.

Such crap deserves to break: better be safe than sorry.

The quotes became necessary only with the advent of LFN-aware file
systems,

NTFS was LFN-aware, from its very beginning, 20+ years ago!

and the Windows API predates the availability of LFN-aware file systems
on Windows.

Which Windows API?

The Win32 API was introduced with Windows NT, together with the LFN-aware
NTFS file system.
It was INCOMPATIBLE to the previously used Win16 API, on source and binary
level.
There was no CreateProcess() in the Win16 API, but WinExec().

So: what about just getting the preconditions to your arguments right?
OUCH^WPooof: ff you do, your arguments just vanish.

If you were the developer of an operating system that millions of people
relied upon and you were faced with the decision between tightening up
the behavior of an API vs. breaking customer applications that people
regularly use, what would your choice be?

I dont need to choose!
There was no compatibility to break.

Stefan

-------- Original message --------
From: Stefan Kanthak
Date:07/30/2014 3:19 AM (GMT-08:00)
To: Michael Cramer , Gynvael Coldwind
Cc: fulldisclosure , Brandon Perry , bugtraq () securityfocus com
Subject: Re: [FD] Beginner's error: import function of Windows Mail executes rogue program C:\Program.exe with 
credentials of other
account


"Michael Cramer" <mike.cramer () outlook com> wrote:

I think you're arguing semantics here.

Of course.

Of course the specifics of how a particular program is executed
will be different between command line and GUI-based OS'.

Really?
Is there any need for this difference you state?
BTW: what is the difference?

Both provide the ability for non-privileged users to launch processes
with specific, temporary Administrative access rather than their
account having always-on Administrative access.

But only Windows' "system call" CreateProcess*() has the COMPLETELY
BRAINDEAD behaviour to guess or probe which executable to run!

Whether or not a user is going to click "yes to everything" is completely
irrelevant to the technical details at hand. It could be argued that
training a user to type 'sudo' on every command could have the same effects.
Just because *YOU* understand the difference doesn't mean there aren't
1000 Linux users out there typing sudo for literally everything simply
because that's how they were taught to type commands in the terminal.

You miss the point, completely.
It's not about "sudo" or RUNAS.EXE or UAC, its about the braindead behaviour
of CreateProcess*().

This is the culprit. Nothing else.

Without its braindead behaviour every developer or his QA would stumble
over unquoted paths the very instant they are used and poorly written
programs with such beginners errors would not be delivered to hundreds
of millions of unsuspecting users putting them at risk to run a rogue
C:\Program.exe



RE: 0.


http://www.dwheeler.com/essays/filenames-in-shell.html


This was brought up on the FD mailing list towards the end of June
(http://seclists.org/fulldisclosure/2014/Jun/140) in which discussion
on how the Linux shell has its own quirks regarding path names and
potential security issues. It's disingenuous to single out Windows with
its path name handling in this context without addressing this issue
across the board. Simply put, these types of issues are universally
understood across many operating systems to be in the realm of
applications themselves and NOT a fault of the platform.

Its but the fault of the platform!
CreateProcess*() is just the "system call", and its the ONLY "system
call" available to start another process, used by ShellExecute*(),
the service manager etc. pp.

So again, to sit here and single out Windows for its quirky pathname
behavior is just picking fights here.

See above: you completely miss the point!

In short, yes, Windows has a function that takes input and launches a
path which COULD lead to unexpected behavior when passing the argument
without quotes. But arguing that this should be on Microsoft to resolve
within Windows just points me back towards the aforementioned link which
says such problems should also be resolved at the OS layer for them as well.

See above: CreateProcess*() is the OS layer here!

To address your Windows auto-elevated processes. Yes, this is the default
configuration when these processes are signed binaries from a specific
Microsoft certificate. However, for the security conscious folks you can
easily turn UAC up to the maximum level which will require elevation on
the secure desktop for all applications, including these processes. This
was the default configuration in Windows Vista when it was released, but
enough people complained about it that Microsoft reduced the # of UAC
prompts in Windows 7 with the above resolution.

Yes. An that's another completely braindead decision.
Instead of fixing their own buggy programs they built one more crappy
workaround.

All user accounts in Ubuntu are created with the ability to use sudo
automatically without further configuration from the root account.

All exec*() calls not only on Ubuntu DONT show the braindead behaviour
of Windows CreateProcess*().

regards
Stefan

Sent from my Surface Pro 3





From: Stefan Kanthak
Sent: ?Monday?, ?July? ?28?, ?2014 ?10?:?43
To: Michael Cramer, Gynvael Coldwind
Cc: fulldisclosure, Brandon Perry, bugtraq () securityfocus com





"Michael Cramer" <mike.cramer () outlook com> wrote:

sudo make-me-a-sandwich.py


How is this different from any other temporary, per-process elevation system?

0. neither sudo nor make-me-a-sandwich.py nor the OS where these programs
  typically run have a CreateProcess*() system call which guesses which
  executable it should run in case of a command line with embedded spaces.

  Do you expect that your command line executes "sudo make-me-a-sandwich.py"
  in the absence of a file sudo or sudo.exe?


1. if you omit sudo from the command line, there is no elevation, not even
  an attempt for an elevation.

  On Windows, you dont need to use sudo, you just "open" for example
  REGEDIT.EXE or make-me-a-sandwich.reg: if you do this in a standard
  user account REGEDIT.EXE will run with standard user rights, without
  any prompt for elevation. But if you do this in an administrator account
  (except the builtin "Administrator"), Windows prompts for consent.

  And if you use one of the 70 Windows programs which Microsoft in their
  very finite wisdom granted auto-elevation, you wont see any elevation
  prompt at all!


2. on *x, your user account is an UNPRIVILEGED user account, and you have
  to use sudo explictly.

  On Windows, all user accounts created during setup are administrator
  accounts which show the above mentioned behaviour.


Is this enough of a difference?

Sent from my Surface Pro 3

ARGH!
I don't need any advertising!

Stefan

From: Stefan Kanthak
Sent: ?Monday?, ?July? ?28?, ?2014 ?06?:?08
To: Gynvael Coldwind
Cc: fulldisclosure, Brandon Perry, bugtraq () securityfocus com





Gynvael Coldwind wrote:

So reading the links you provided I semi-agree with you. I think the
problem boils down to this part of your initial e-mail:

PS: yes, it needs administrative privileges to write C:\Program.exe.
   BUT: all the user account(s) created during Windows setup have
   administrative privileges.

My point was (and it still stands) that if you have admin access, this
isn't a privilege escalation, as there is no "escalation" part here.

Correct.
If only Microsoft would educate its users to exercise STRICT user
separation and use different accounts for administration and daily work.

This is where and why UAC chimes in (which answers your question below):
Joe Average uses the administrative account created during Windows setup,
but UAC strips the administrator rights.
Microsoft "sells" UAC as "Joe Average works with standard user rights"
or "Joe Average is not an administrator any more", neglecting that Joe
will happily approve almost every request for administrative rights (or
isnt asked at all when one of the about 70 Windows executables which are
exempt from the elevation prompt are auto-elevated).

The links you provided use different wording, e.g.
(http://blogs.technet.com/b/srd/archive/2013/07/09/assessing-risk-for-the-july-2013-security-updates.aspx):
"To exploit the vulnerability addressed by this update, attacker must
have permission to create a new file at the root of the system drive.
(C:\malicious.exe)"

This makes of course more sense, though as I did mention above, it
does seem to require deliberate action from the administrator to
actually allow a non-admin user the WD (add file to directory)
privilege on C:\, which is rather rare I would say.

Correct.
This argument holds as long as strict user separation is exercised.
But with UAC, Joe Average is both user and administrator, and isnt
really aware of his split personality.

That being said, after thinking about it again I do see your point,
which I interpret at: even if an administrator grants all users WD/AD
on C:\, there should be no reason for him to worry, as there is no
reason to suspect files placed in C:\ are going to auto-execute on
certain events*.
* let's leave autoexec.bat/config.sys out of this, as that branch of
Windows is long dead and supported only FAT anyway

So let me change my initial e-mail to: Congratz on finding the bug :)

(BTW not sure why did you bring UAC into the discussion - did I miss
something? or was it just an argument you've heard before and wanted
to reply to it preventively?)

Cheers!

regards
Stefan


On Fri, Jul 25, 2014 at 2:50 PM, Stefan Kanthak <stefan.kanthak () nexgo de> wrote:
Gynvael Coldwind wrote:

Well it was discussed a couple of times recently on FD that this is a bug,
but it's not a privilege escalation.
If you are admin (and you did mention that it's a prerequisite) you can
execute code as other users anyway - so there's no *escalation* here.

Therefore it's not a security bug (unless you are using a super old version
of Windows with incorrect ACLs on c:\, which sounds like a bug in itself),
just a "normal" bug.
Not sure if FD is the right place for non-security bugs tbh.

If these bugs were no security bugs: why does Microsoft then publish fixes
for (at least some of) them via MSRC bulletins and Windows Update?

See <https://technet.microsoft.com/library/security/ms13-058.aspx>
or <https://technet.microsoft.com/library/security/ms13-034.aspx>

Or pulls drivers whose setup routines show these bugs from Windows Update?

See <http://seclists.org/fulldisclosure/2014/May/40>


Also try to see these bugs as a blended threat:

* during Windows setup Microsoft still creates all user accounts as
  administrators.

* Microsoft sells its unsuspecting users UAC as a security feature, but does
  NOT inform them (or at least does not inform Joe Average) that UAC is not
  a security boundary and they should better use a restricted^Wstandard user
  account instead of the administrator account created during setup.

* Joe Average will happily give consent to any program which presents an UAC
  prompt to him: he wants to get his work done, and this UAC prompt is just
  an annoyance. BTW: when Windows asks him for consent, this must be right?

regards
Stefan

Cheers,
On 25 Jul 2014 00:46, "Stefan Kanthak" <stefan.kanthak () nexgo de> wrote:

Brandon Perry wrote:

So, I am very curious how you are finding these? Have you automated this
or
is it manual hand work?

All my Windows installations have
<http://home.arcor.de/skanthak/download/SENTINEL.EXE> and
<http://home.arcor.de/skanthak/download/SENTINEL.DLL> preinstalled as
C:\Program.exe and C:\Program.dll, so I'm notified when some poorly written
program tries to execute them.
The sentinels call MessageBox() with "MB_SERVICE_NOTIFICATION", so the
messages are recorded in the event log too where I can find them later.

I also preinstall an APPINIT.DLL <https://support.microsoft.com/kb/197571>
which logs all command lines of programs linked to USER32.DLL to a file:
filtering for "C:\Program " at column 1 lists all the culprits.

My third source is a SAFER.Log <
https://technet.microsoft.com/cc786941.aspx>
where every execution attempt is logged, including the executables caller:
filtering this for "\program.exe" or "\program.dll" lists all the culprits.

So basically I just have to sit and wait...

In case one of my customers was hit, and this did not happen during an
installation, I have to interrogate them what they did... and hope they can
remember with sufficient detail.

But almost all hits occur during installations or the customization
following
an installation (here it was the import of existing mails into a new
account),
so these are not so difficult to reproduce.

regards
Stefan

PS: of course it helps if 8.3 names are disabled and "C:\Program Files\"
can't
    be aliased as C:\Progra~1\
    To achieve this just run FORMAT C: /FS:NTFS /S:Disable in Windows PE
    before you start the installation of Windows 7 and later.
    For Windows NT5.x you'll have to use \i386\MIGRATE.INF

On Wed, Jul 23, 2014 at 2:50 PM, Stefan Kanthak <stefan.kanthak () nexgo de

wrote:

Hi @ll,

the import function of Windows Mail executes a rogue program
C:\Program.exe
with the credentials of another account, resulting in a privilege
escalation!

[...]

_______________________________________________
Sent through the Full Disclosure mailing list
http://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: http://seclists.org/fulldisclosure/





-- 
Gynvael Coldwind


  By Date           By Thread  

Current thread:
[ Nmap | Sec Tools | Mailing Lists | Site News | About/Contact | Advertising | Privacy ]