Re: [PATCH 2/2] groups: Allow unprivileged processes to use setgroups to drop groups

From: Eric W.Biederman
Date: Mon Nov 17 2014 - 17:45:41 EST




On November 17, 2014 1:46:59 PM EST, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>On Mon, Nov 17, 2014 at 10:31 AM, Andy Lutomirski <luto@xxxxxxxxxxxxxx>
>wrote:
>> On Mon, Nov 17, 2014 at 10:06 AM, Casey Schaufler
>> <casey@xxxxxxxxxxxxxxxx> wrote:
>>> On 11/15/2014 1:01 AM, Josh Triplett wrote:
>>>> Currently, unprivileged processes (without CAP_SETGID) cannot call
>>>> setgroups at all. In particular, processes with a set of
>supplementary
>>>> groups cannot further drop permissions without obtaining elevated
>>>> permissions first.
>>>
>>> Has anyone put any thought into how this will interact with
>>> POSIX ACLs? I don't see that anywhere in the discussion.
>>
>> That means that user namespaces are a problem, too, and we need to
>fix
>> it. Or we should add some control to turn unprivileged user
>namespace
>> creation on and off and document that turning it on defeats POSIX
>ACLs
>> with a group entry that is more restrictive than the other entry.
>>
>
>This is a significant enough issue that I posted it to oss-security:
>
>http://www.openwall.com/lists/oss-security/2014/11/17/19
>
>It's not at all obvious to me how to fix it. We could disallow userns
>creation of any supplementary groups don't match fsuid, or we could
>keep negative-only groups around in the userns.
>
>It may be worth adding a sysctl to change the behavior, too. IMO it's
>absurd to use groups to deny permissions that are otherwise available.

There is an obvious user namespace fix. Don't allow dropping supplemental groups that are not mapped.

That will require a little bit of fancy footwork if you want to play with supplemental groups in your unprivileged user namespace. I would like to get a grip on what hoops would be required before we add the additional restriction. Possibly something as simple as calling sg.

I also want to look at what Tizen and any other concrete pieces of code I can find using this negative permission pattern are actually doing. Bugs definitely exist, but I have this erie feeling that the bugs may be in instances of userspace using this negative group permission pattern. I think we may have a hideous case of one setuid binary defeating a privilege check of another piece of code.

This issue looks like it is worth a full scale investigation. Sigh.

Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/