On Fri, 28 Sep 2012 at 14:30:37 -0700, Geoffrey Thomas wrote: > CVE-2012-3524 is about setuid binaries linking libdbus being easily > trickable to do bad things via a malicious PATH (for finding > dbus-launch), or through a DBUS_* address variable using the > unixexec address type.
This also affects reimplementations of D-Bus, most prominently GDBus in GLib; src:dbus can't do anything to help those. I believe the libdbus part of this CVE affects wheezy and certain unusual squeeze configurations. The known vectors for privilege escalation are: * tell libdbus (explicitly or via it being the session-bus default) to connect to autolaunch: which results in it exec'ing dbus-launch; have it not find dbus-launch in its configured ${bindir}; have it find a malicious dbus-launch substitute in its $PATH * tell libdbus to connect to unixexec:something-malicious The former is only exploitable if there is no ${bindir}/dbus-launch. On some distributions, D-Bus is configured with ${prefix} = /, but dbus-launch is moved to /usr/bin by the packager because it depends on libX11. As a result, libdbus tries /bin/dbus-launch, never finds it, and falls back to searching the $PATH, where it hopefully finds /usr/bin/dbus-launch. However, on Debian, we do the opposite: configure D-Bus with ${prefix} = /usr, and move libdbus-1.so.3 into /lib in the packaging. As a result, the only way to not find dbus-launch in /usr/bin is if you have libdbus-1-3 but not dbus-x11. This mitigates the attack somewhat. > Initially the D-Bus developers thought that > this should be fixed on the application side (hence the comment in > the security-tracker), but decided that it would be better to have a > defense-in-depth approach, and change _dbus_getenv to not succeed if > the current program is setuid or similar, since that's faster than > patching every relevant program. I still think this is an application bug - it's the application that knows it is (or claims to be) setuid-safe - but yes, we should do both. S -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org