oss-security - CVE-2014-8156: freesmartphone.org stack configures D-Bus system bus to be insecure

Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [day] [month] [year] [list]
Date: Tue, 27 Jan 2015 21:29:16 +0000
From: Simon McVittie <simon.mcvittie@...labora.co.uk>
To: oss-security@...ts.openwall.com
Subject: CVE-2014-8156: freesmartphone.org stack configures D-Bus system bus
 to be insecure

Type of vulnerability: CWE-284 Improper Access Control
Exploitable by: local users
Impact: unknown, dependent on installed D-Bus system services; at least
local denial of service
Reporter: Simon McVittie, Collabora Ltd.
Reported to vendor: 2015-01-13

While checking Debian for incorrect/dangerous D-Bus security policy
files (found in /etc/dbus-1/system.d/*.conf) I found that the
freesmartphone.org (fso) stack contains several problematic rules.

Summary of modules confirmed to be affected:

Debian package fso-gsmd version 0.12.0-3
Debian package fso-usaged version 0.12.0-2
Upstream cornucopia.git (fsoaudiod, fsodatad, fsodeviced, fsogsmd,
fsonetworkd, fsotdld, fsousaged) git master on 2015-01-19

Debian package fso-frameworkd version 0.9.5.9+git20110512-4
Upstream framework.git version 0.10.1 and git master on 2015-01-19

In addition, Debian packaging for one fso-related package, also
available in Ubuntu and possibly other derivatives, has a similar flaw:

Debian package phonefsod version 0.1+git20121018-1
(Upstream code for phonefsod
<http://git.shr-project.org/git/?p=phonefsod.git> was not vulnerable as
of 2015-01-19)

Other fso modules might be affected, I only scanned the ones available
in Debian.

Please see below for further technical details of the faulty security
policies, and some advice on fixing them, which is applicable to any
project with D-Bus security policy files. For further advice, please
contact dbus@...ts.freedesktop.org (public mailing list) or
dbus-security@...ts.freedesktop.org (non-public list for embargoed
security vulnerabilities), and explain what security policy you intend
your service to have.

Category 1: send_path != "/"
----------------------------

This refers to configurations like this:

    <policy context="default">
        ...
        <allow send_path="/org/freesmartphone/Framework"/>

This allows every local user to send arbitrary D-Bus messages to the
path /org/freesmartphone/Framework on *any* D-Bus system service (rough
HTTP analogy: send a POST to http://server/org/freesmartphone/Framework
on any server).

At first glance, this might seem harmless, because why would a non-fso
service have any functionality at that path? However, services that use
a low-level binding like libdbus and implement their functionality via a
message filter do not necessarily differentiate between paths.

Notably, org.freedesktop.DBus, the pseudo-service provided by
dbus-daemon, provides the same API at every object path (and is
constrained to do so by backwards compatibility). dbus releases 1.8.14
and 1.9.6 mitigate this by locking down functionality that is not
intended to be public, so that it can only be accessed at the canonical
path /org/freedesktop/DBus. However, in earlier dbus releases, this can
be used to make the dbus-daemon consume arbitrary amounts of memory.

It is possible that other system services have similar behaviour; the
worst-case impact is arbitrary root code execution, although I do not
know of a specific service where this would happen.

Example files: etc/dbus-1/system.d/frameworkd.conf in fso-frameworkd,
data/fsogsmd.conf in fso-gsmd, data/fsousaged.conf in fso-usaged

In addition, while data/dbus-1/phonefsod.conf in
phonefsod/0.1+git20121018-1 is not vulnerable, the Debian-specific
version in debian/phonefsod.conf has the same issue.

Category 2: send_path = "/"
---------------------------

This is a variation of category 1 where the affected path is "/":

    <policy context="default">
        ...
        <allow send_path="/"/>

This is potentially more serious than category 1, because it affects any
system service that either provides the same API on every object path,
or specifically provides API at "/". Older versions of BlueZ are one
example of a service that specifically uses "/".

etc/dbus-1/system.d/frameworkd.conf in
fso-frameworkd/0.9.5.9+git20110512-4 has this pattern.

Fixing these vulnerabilities
----------------------------

Where possible, the recommended pattern is for the only <allow> rules to
be <allow own="x.y.z"/> and/or <allow send_destination="x.y.z"/>.

If it is necessary to control different object paths differently, either
use Polkit/PolicyKit, or specify both in the same <allow> rule:

    <allow send_destination="x.y.z" send_path="/a/b/c"/>

This is treated as an "logical and" operation: the rule will only match
sending messages that meet all the criteria. Similarly, you can combine
send_destination with send_interface and/or send_member in the same
<allow> or <deny> rule.

The best practice is that every rule with <allow send_something> should
have a send_destination attribute.

If the destination will not own a well-known bus name (e.g. the "agent"
pattern in which BlueZ calls out to agent processes running as GUI
users) and so send_destination cannot be used, prefer <allow
send_interface> instead of <allow send_path>, and only allow root (or a
specific daemon user, if not root) to do this. For instance, this is
acceptable:

    <policy user="root">
        <allow send_interface="com.example.MyAgent"/>
    </policy>

Parts of the fso D-Bus policy files indicate a misunderstanding of these
files' syntax. If you write multiple <allow send_*> rules in the same
<policy> block, that is effectively an "logical or" operation: each rule
separately allows sending messages that match it.

In particular, there is no functional difference between

    <policy context="default">
        <allow send_path="/x/y/z"/>
        <allow send_destination="x.y.z"/>
    </policy>

and

    <policy context="default">
        <allow send_path="/x/y/z"/>
    </policy>
    <policy context="default">
        <allow send_destination="x.y.z"/>
    </policy>

and the send_path rule is equally problematic in both spellings. The
<policy> only controls who the rules apply to (everyone/a specific
user/a specific group); it does not provide "logical and" semantics.

    S
Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.