[<prev] [next>] [day] [month] [year] [list]
Message-ID: <Pine.LNX.4.44L.0303192008160.30372-100000@kuku.eu.org>
Date: Wed, 19 Mar 2003 20:22:45 +0100 (CET)
From: Andrzej Szombierski <qq@...u.eu.org>
To: bugtraq@...urityfocus.com
Subject: linux kmod/ptrace bug - details
Hello
There are many discussions (on slashdot for example) on the recent linux
ptrace (& kmod) bug. I'll try to clarify what is this all about.
It's a local root vulnerability. It's exploitable only if:
1. the kernel is built with modules and kernel module loader enabled
and
2. /proc/sys/kernel/modprobe contains the path to some valid executable
and
3. ptrace() calls are not blocked
These conditions are met on most standard linux distros.
Ok now how it works:
When a process requests a feature which is in a module, the kernel spawns
a child process, sets its euid and egid to 0 and calls execve("/sbin/modprobe")
The problem is that before the euid change the child process can be
attached to with ptrace(). Game over, the user can insert any code into a
process which will be run with the superuser privileges.
Solutions/workarounds:
- patch the kernel
or
- disable kmod/modules
or
- install a ptrace-blocking module
or
- set /proc/sys/kernel/modprobe to /any/bogus/file
A word about 2.5. kernels - these are not vulnerable because the kernel
thread spawning code has been rewritten so that the modprobe process is
spawned from keventd, it never runs with non-root uid, so it can't be
ptraced by any non-root user.
Sample exploit here (ix86-only):
http://august.v-lo.krakow.pl/~anszom/km3.c
--
: Andrzej Szombierski : anszom@...o.krakow.pl : qq@...u.eu.org :
: anszom@...kitu.com ::: radio bez kitu <=> http://bezkitu.com :
Powered by blists - more mailing lists