Re: CVE Request: libpam-sshauth: local root privilege escalation

[Scott Balneaves <sbalneav () ltsp org>]

> > Here, the commit message for revision 93 was "Succeed for system
> > accounts."
> >
> > We don't know why introducing the undocumented behavior of "Is it a
> > system user? Fail" would be better than simply not checking
> > "pwent->pw_uid < UID_MIN" at all. Also, is there any risk that, with
> > this libpam-sshauth update, a system's PAM configuration might
> > suddenly provide no way for root to login via SSH?
> >
> > Is it possible that the original motivation for revision 93 was that
> > the PAM_SUCCESS from pam_sm_authenticate was supposed to be specially
> > handled elsewhere in the "pwent->pw_uid < UID_MIN" case?

The problem was, quite bluntly, an incomplete understanding of PAM
mechanics on my part.

The original idea was that it was supposed to be used in conjunction
with other modules; specifically, pam_unix.  So my *thinking* (if you
could call it that) was that it would be used as such:

auth required pam_unix.so  ...
auth required pam_sshauth.so ...

Since (in my mind), accepting the root user would be handled by pam_unix,
I should simply succeed, since if the root account password was
incorrectly entered, the pam_unix result would be a FAIL, and thus
then entire pam auth stack would fail.  Therefore, in my (incorrect)
thinking,
I should simply succeed on systems accounts.

I didn't, at the time, know about the ability to skip with [success=N], or
even
consider that I would use it as the only pam module.

TL;DR: I didn't know what I was doing, and misunderstood how I should
handle systems accounts.  Mea culpa, mea maxima culpa.

Cheers,
Scott

On Tue, May 3, 2016 at 1:51 PM, Vagrant Cascadian <vagrant () debian org>
wrote:

> On 2016-05-03, Salvatore Bonaccorso wrote:
> > On Sun, May 01, 2016 at 10:02:15AM -0400, cve-assign () mitre org wrote:
> > > > Due to a programming error, libpam-sshauth returned PAM_SUCCESS where
> > > > it should fail with PAM_AUTH_ERR. This was fixed in Debian in the last
> > > > upload to unstable with the attached patch.
> https://bazaar.launchpad.net/~ltsp-upstream/ltsp/libpam-sshauth/revision/114
> > > We can assign a CVE ID because it appears that something definitely is
> > > wrong from the Debian perspective, either the code itself or
> > > documentation/lack-of-documentation about how the code was supposed to
> > > be used.
> > >
> > > Use CVE-2016-4422.
> >
> > Thanks for assigning the CVE identifier.
> >
> > > However, we don't completely understand the issue:
> > >
> > > > Introduced with:
> https://bazaar.launchpad.net/~ltsp-upstream/ltsp/libpam-sshauth/revision/93/src/pam_sshauth.c
> > > Here, the commit message for revision 93 was "Succeed for system
> > > accounts."
> > >
> > > We don't know why introducing the undocumented behavior of "Is it a
> > > system user? Fail" would be better than simply not checking
> > > "pwent->pw_uid < UID_MIN" at all. Also, is there any risk that, with
> > > this libpam-sshauth update, a system's PAM configuration might
> > > suddenly provide no way for root to login via SSH?
> > >
> > > Is it possible that the original motivation for revision 93 was that
> > > the PAM_SUCCESS from pam_sm_authenticate was supposed to be specially
> > > handled elsewhere in the "pwent->pw_uid < UID_MIN" case?
> > >
> > > Although not directly applicable to libpam-sshauth, the examples
> > > section of the
> > > http://www.linux-pam.org/Linux-PAM-html/sag-pam_succeed_if.html man
> > > page shows that a set of rules is sometimes designed with UID_MIN
> > > special cases.
> >
> > It might be right that revision 93 cannot be considred the introducing
> > revision for the problem. By following the example as given in the
> > README.
> >
> > https://sources.debian.net/src/libpam-sshauth/0.3.1-1/README/#L75
> >
> > $ cat /etc/pam.d/testservice
> > auth    required        pam_sshauth.so host=127.0.0.1 nostrict # or
> wherever
> > auth    required        pam_exec.so expose_authtok /usr/bin/ltsp-session
> > session required        pam_exec.so /usr/bin/ltsp-session
> > $ pamtester -v testservice root authenticate open_session close_session
> > pamtester: invoking pam_start(testservice, root, ...)
> > pamtester: performing operation - authenticate
> > Password: <anypassword>
> > pamtester: successfully authenticated
> > pamtester: performing operation - open_session
> > pamtester: successfully opened a session
> > pamtester: performing operation - close_session
> > pamtester: session has successfully been closed.
> >
> > I want though to add the Debian maintainer for libpam-sshauth to more
> > accurately answer the raised questions, Vagrant Cascadian
> > <vagrant () debian org>.
>
> Also bringing the primary upstream developer, Scott Balneaves
> <sbalneav () ltsp org> into the conversation, who has better understanding
> of the code.
>
> For this issue, I've largely just discovered it and made some small
> effort to backport the patch.
>
>
> live well,
>   vagrant