linux-kernel - Re: [PATCH] proc: fix NULL dereference when reading /proc/<pid>/auxv

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20161020124405.GL14609@dhcp22.suse.cz>
Date:   Thu, 20 Oct 2016 14:44:06 +0200
From:   Michal Hocko <mhocko@...nel.org>
To:     Leon Yu <chianglungyu@...il.com>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Al Viro <viro@...iv.linux.org.uk>,
        Kees Cook <keescook@...omium.org>,
        Oleg Nesterov <oleg@...hat.com>,
        John Stultz <john.stultz@...aro.org>,
        Mateusz Guzik <mguzik@...hat.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Janis Danisevskis <jdanis@...gle.com>,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] proc: fix NULL dereference when reading /proc/<pid>/auxv

On Thu 20-10-16 20:32:43, Leon Yu wrote:
> On Thu, Oct 20, 2016 at 1:17 AM, Michal Hocko <mhocko@...nel.org> wrote:
> > So here is my RFC as an alternative. Thoughts? Please note that we
> > currently have only very few users of use_mm() API in the kernel
> > so a risk of a regression is not really high. usb/gadget are using it
> > only temporarily. The remaining is vhost which operates on a remote mm
> > and I have no idea whether somebody might abuse /proc/vhost/mem or
> > anything - let's add Michael to the CC list. I am pretty sure nobody
> > abuse oom_reaper proc directory as this one is pretty new and such a
> > usage would be pretty much undefined as the reaper unmaps the address
> > space.
> 
> With this patch I cannot tell the difference between a) the thread is
> exiting and b) it's a kernel thread,
> besides, getting "no such process" while the kthread does exist is a
> bit confusing.

Do we really need to distinguish those two cases? In other words under
which conditions something would fail when seeing a pid directory and
ESRCH when opening a file?

> IMO, reading /proc/<kthread_pid>/auxv and getting empty output are
> quite straightforward,
> it doesn't seem to be that "abusive".

Then we need to teach all those implementations to check for kthread
explicitly and return an empty output rather than relying on mm == NULL
because that, as explained, might belong to some process.

-- 
Michal Hocko
SUSE Labs