oss-sec mailing list archives

Re: [Pkg-xfce-devel] Bug#639151: Bug#639151: Bug#639151: Local privilege escalation


From: Sebastian Krahmer <krahmer () suse de>
Date: Fri, 26 Aug 2011 10:43:55 +0200


Hi,

You probably dont take into account the chown() that happens in lightdm.
Just unlink the created ~/.dmrc or ~/.Xauthority files after creation and make a symlink
to /etc/passwd to chown it to yourself.
However I didnt dig deep enough into it to write an exploit as I dont have
a working lightdm setup. The correct behavior is to temporarily drop euid/fsuid
to that of the user if doing anything with his files.

The PAM issue that I was curious about was that a pam_start() etc is done
for the greeter-user (which I expect to be some "lightdm" user)?

I would expect all pam_ calls are only done for the user who is actually
about to login. The question that came up to me was whether pam_environment
from the user would have impact on uid-0 called programs/scripts since
you transfer the PAM env to the process env.

Sebastian

On Thu, Aug 25, 2011 at 05:54:23PM +0200, Yves-Alexis Perez wrote:
On mer., 2011-08-24 at 20:55 +0200, Yves-Alexis Perez wrote:
And, out of curiosity, how would you achieve privilege escalation? You
should be able to erase/rewrite arbitrary files, including /etc/shadow,
but you don't really have control on what's written there. 

In gdm (CVE-2011-0727 I guess) the issue was that a g_file_copy() was
run as root from files under user control (.dmrc and the avatar), to a
cache dir with write permissions (afaict). So it was easy to put
whatever stuff you need in the original file and make a symlink
to /etc/shadow in the destination folder so the g_file_copy() would
erase that:

                 res = g_file_copy (src_file,
                                    dst_file,
                                    G_FILE_COPY_OVERWRITE |
                                    G_FILE_COPY_NOFOLLOW_SYMLINKS,
                                    NULL,
                                    NULL,
                                    NULL,
                                    &error);


I'm not too sure what G_FILE_COPY_OVERWRITE means, if it truncate()s and
write over of if it unlink()s and start fresh (digging in glib to find
out). Apparenlty in the fallback case (not sure if it's the case here)
it ends up doing a g_file_replace()).

In any case, in lightdm case, for .Xauthority file it uses
g_file_replace() which creates a temporary file and then rename over the
new file, so in the worst case you overwrite a system file with
xauthority data.

Same thing for .dmrc, you can overwrite system files but with dmrc data
which look like 

[Desktop]
Session=xfce
Lang=fr_FR.UTF-8

so it doesn't look easy to gain root access with that.

LightDM maintains a cache for dmrc files in /var/cache/lightdm but the
folder is created 0700 so it doesn't look like one can put symlinks
there and have it use a user-controled .dmrc.

All in all, I'm not too sure there's a privilege escalation for
Xauthority/.dmrc files (but if one exists, I'm interested in how to do
it, by curiosity). But you still damage pretty much any arbitrary file,
which is still an easy DoS.

Regards,
-- 
Yves-Alexis



-- 

~ perl self.pl
~ $_='print"\$_=\47$_\47;eval"';eval
~ krahmer () suse de - SuSE Security Team

---
SUSE LINUX Products GmbH,
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer, HRB 16746 (AG Nürnberg)
Maxfeldstraße 5
90409 Nürnberg
Germany


Current thread: