Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 14 Feb 2014 23:07:59 +0100
From: Helmut Grohne <>
Subject: Re: Re: Bug#738855: initscripts: Skip killing
 root-owned process starting with @

Sorry for being unclar here.

On Fri, Feb 14, 2014 at 11:50:29AM -0500, wrote:
> >
> > Message #34
> This message starts by discussing initscripts, but ends by discussing
> a CVE assignment for systemd. That CVE assignment would potentially be
> reasonable, but we wanted to first clarify what is being asked. We
> think you mean:
>   - adding a patch to initscripts to introduce more compatibility
>     between initscripts and systemd may be considered a security
>     enhancement, and probably would not be considered a vulnerability
>     fix, so no CVE ID is being requested for a problem in the
>     unpatched initscripts code

Almost. The patch proposed for initscripts does not enhance its
security, but it makes it easier to write daemons against both systemd
and sysv init (initscripts). The anticipated problem is not present in
initscripts, but it could be introduced via the compatibility patch.
Whether there actually is a vulnerability in the patch or not, nothing
has been shipped to actual systems and that is why no CVE is being
requested here.

>   - this systemd commit
>     introduced the killall.c file. In the first version of this file, the
>        /* Non-root processes otherwise are always subject to be killed */
>        if (uid != 0)
>                return false;
>        ...
>        /* Processes with argv[0][0] = '@' we ignore from the killing
>         * spree.
>         *
>         * */
>         if (count == 1 && c == '@')
>                 return true;
>         return false;
>     code was included.
>   - you are proposing that the above "return true" line is a
>     vulnerability because it may allow a not-fully-privileged root
>     user to cause data loss. This could possibly have one CVE ID. One
>     of the arguments against assigning a CVE ID is that this "return
>     true" could have been an intentional tradeoff between perfect
>     privilege checking and design complexity. For environments with
>     not-fully-privileged root users, we're not sure whether there's
>     general acceptable of a guideline that OS components must never
>     contain any program logic to make any security-relevant decision
>     on the basis of the uid value.

You have well captured the intended message. Fundamentally the issue
boils to the question of whether there is such a thing as an
"unprivileged process with effective UID 0". If there isn't, then there
is no vulnerability in systemd (and the proposed patch to initscripts is
fine as well). Specifically, I am seeking input on whether there are
practical situations in which a process with effective UID 0 can be
considered less privileged than say accessing arbitrary block devices
for writing (which would have similar implications wrt. data loss). The
anticipated scope is LSMs and Linux capabilities in action.

At least theoretically, a process could be jailed in a chroot with most
of its capabilities (especially CAP_MKNOD and CAP_CHROOT) revoked.  Such
a process could be considered unprivileged. The state of the art to
craft such a jail would more likely involve Linux user namespaces though
and thus avoid effective UID 0 alltogether.

>   - versions of systemd before
>     bd3fa1d2434aa28564251ac4da34d01537de8c4b, in which killall.c did
>     not exist, may have had other problems because the right processes
>     were not killed at the right times. This could possibly have a
>     second CVE ID if there were security implications.

I am making no claims about systemd's behaviour before this commit. In
any case those early versions tend not to be around in actual systems,
so I doubt that tracking vulnerabilities there would be useful at


Powered by blists - more mailing lists

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.