linux-kernel - Re: shmctl(SHM_STAT) vs. /proc/sysvipc/shm permissions discrepancies

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] [day] [month] [year] [list]

Date:   Mon, 12 Feb 2018 09:30:37 -0800
From:   Davidlohr Bueso <dave@...olabs.net>
To:     "Michael Kerrisk (man-pages)" <mtk.manpages@...il.com>
Cc:     Michal Hocko <mhocko@...nel.org>,
        Linux API <linux-api@...r.kernel.org>,
        Manfred Spraul <manfred@...orfullife.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Al Viro <viro@...iv.linux.org.uk>,
        Kees Cook <keescook@...omium.org>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Mike Waychison <mikew@...gle.com>,
        LKML <linux-kernel@...r.kernel.org>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>
Subject: Re: shmctl(SHM_STAT) vs. /proc/sysvipc/shm permissions discrepancies

On Thu, 21 Dec 2017, Michael Kerrisk (man-pages) wrote:

>Hi Michal,
>
>On 21 December 2017 at 09:02, Michal Hocko <mhocko@...nel.org> wrote:
>> On Wed 20-12-17 17:17:46, Michael Kerrisk wrote:
>>> Hello Michal,
>>>
>>> On 20 December 2017 at 10:20, Michal Hocko <mhocko@...nel.org> wrote:
>>> > On Tue 19-12-17 17:45:40, Michael Kerrisk wrote:
>>> >> But, is
>>> >> there a pressing reason to make the change? (Okay, I guess iterating
>>> >> using *_STAT is nicer than parsing /proc/sysvipc/*.)
>>> >
>>> > The reporter of this issue claims that "Reading /proc/sysvipc/shm is way
>>> > slower than executing the system call." I haven't checked that but I can
>>> > imagine that /proc/sysvipc/shm can take quite some time when there are
>>> > _many_ segments registered.
>>>
>>> Yes, that makes sense.
>>>
>>> > So they would like to use the syscall but
>>> > the interacting parties do not have compatible permissions.
>>>
>>> So, I don't think there is any security issue, since the same info is
>>> available in /proc/sysvipc/*.
>>
>> Well, I am not sure this is a valid argument (maybe I just misread your
>> statement).
>
>(Or perhaps I was not clear enough; see below)
>
>> Our security model _might_ be broken because of the sysipc
>> proc interface existance already. I am not saying it is broken because
>> I cannot see an attack vector based solely on the metadata information
>> knowledge. An attacker still cannot see/modify the real data. But maybe
>> there are some bugs lurking there and knowing the metadata might help to
>> exploit them. I dunno.
>>
>> You are certainly right that modifying/adding STAT flag to comply with
>> the proc interface permission model will not make the system any more
>> vulnerable, though.
>
>Yep, that was my point. Modifying _STAT behavior won't decrease security.
>
>That said, /proc/sysvipc/* has been around for a long time now, and
>nothing bad seems to have happened so far, AFAIK.
>
>>> The only question would be whether
>>> change in the *_STAT behavior might surprise some applications into
>>> behaving differently. I presume the chances of that are low, but if it
>>> was a concert, one could add new shmctl/msgctl/semctl *_STAT_ALL (or
>>> some such) operations that have the desired behavior.
>>
>> I would lean towards _STAT_ALL because this is Linux specific behavior
>> (I have looked at what BSD does here and they are checking permissions
>> for STAT as well). It would also be simpler to revert if we ever find
>> that this is a leak with security consequences.

So I took a crack at this, and my only doubt is whether or not the lsm
security hooks should be considered or not. Specifically, should the
SHM_STAT_ALL case consider security_shm_shmctl()?

While the relevant persmission checks that allow for the discripancies
between 0444 procfs and a 0600 via creating the ipc object are done in
ipcperms() returning -1, is there a scenario where some lsm policy could
change the /proc/sysvipc/ interface? If not, I think we can avoid it, but
I'm not a security person.

Thanks,
Davidlohr

>
>Oh -- I was unaware of this BSD behavior. At least on the various UNIX
>systems that I ever used SYSVIPC (including one or two ancient
>commercial BSD derivatives), ipcs(1) showed all IPC objects. (On
>FeeBSD, at least, it looks like ipcs(1) doesn't use the *_STAT
>interfaces.)
>
>Cheers,
>
>Michael
>
>
>
>-- 
>Michael Kerrisk
>Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
>Linux/UNIX System Programming Training: http://man7.org/training/