Re: CVE request: kernel: taskstats/procfs io infoleak

[Eugene Teo <eugeneteo () kernel org>]

On 06/24/2011 08:34 PM, Vasiliy Kulikov wrote:
> Hi,
>
> On Tue, Jun 21, 2011 at 15:24 -0400, Josh Bressers wrote:
> > > /*
> > > * This program tries to learn whether ~user/.ssh/authorized_keys exists
> > > * and is nonempty for any user on local machine. It uses world-readable
> > > * taskstats' nature to get somewhat private io statistics information.  If
> > > * implant taskstats or /proc//io polling into ssh client, it would be
> > > * possible to learn precise authorized_keys' size (and estimate private
> > > * key's(s') size).
> >
> > Are you considering this a flaw, or just an interesting security exercise?
> > Nothing currently comes to mind, but it's possible there could be other
> > data where knowing it exists and the size would be useful.
>
> It can be used to learn ssh and ftp password length.  If privsep is
> enabled in openssh and vsftpd, the unprivileged process' activity very
> precisely shows password information.
>
> For vsftpd read characters count is strlen("USER username\r\n") +
> strlen("PASSWD pass\r\n") + 1, where 1 is one byte read from a pipe
> related to a privileged parent.  If measure statistics between user and
> passwords commands, actual password length and username length can be
> gathered.
>
> For ssh, vice versa, networking activity is constant in packets length,
> but interprocess communications, specifically passwords, depend on
> user input.
>
> For ssh pass_len = wchars - CONST, for vsftpd pass_len = rchars - CONST.
>
> Another daemons with more or less constant io activity might be
> vulnerable too.  PAM greatly complicates precise measurements.
>
>
> I think it needs 2 CVE, one for /proc/PID/io and another for taskstats.
>
> https://lkml.org/lkml/2011/6/24/88

procfs io - CVE-2011-2495
taskstats - CVE-2011-2494

Thanks, Eugene