Re: [PATCH] namei: permit linking with CAP_FOWNER in userns

From: Andy Lutomirski <luto@amacapital.net>
To: Serge Hallyn <serge.hallyn@ubuntu.com>
Cc: Dirk Steinmetz <public@rsjtdrjgfuzkfg.com>,
	Seth Forshee <seth.forshee@canonical.com>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	Linux FS Devel <linux-fsdevel@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	Kees Cook <keescook@chromium.org>,
	Serge Hallyn <serge.hallyn@canonical.com>
Subject: Re: [PATCH] namei: permit linking with CAP_FOWNER in userns
Date: Mon, 2 Nov 2015 11:57:26 -0800	[thread overview]
Message-ID: <CALCETrUPyFMAk9uxJ8+hg0YoMN=mTM3q9jAT7zki1g9uuWtVkg@mail.gmail.com> (raw)
In-Reply-To: <20151102180252.GD1822@ubuntumail>

On Mon, Nov 2, 2015 at 10:02 AM, Serge Hallyn <serge.hallyn@ubuntu.com> wrote:
> Quoting Dirk Steinmetz (public@rsjtdrjgfuzkfg.com):
>>
>> > We've already dealt with such regressions and iirc agreed that they were
>> > worthwhile.
>> Would you prefer to not fix the issue at all, then? Or would you prefer to
>
> No.  I think I was saying I think it's worth adding the 'gid must be mapped'
> requirement.
>
> And I was saying that changing the capability needed is not ok.
>
>> add a new value on /proc/sys/fs/protected_hardlinks -- which would still
>> cause the symptoms you describe on distributions using the new value, but
>> would be more easy to change for users knowing that this is an issue?
>>
>> I personally still favor changing the behavior and documentation over a new
>> value there, as -- once identified by the user -- the user can easily adapt
>
> I agree.
>
> Note the difference - changing the capability required to link the
> file can affect (probably rare, but definately) normal, non-user-namespace
> setups.  Changing the link requirements in a user namespace so that gid
> must be mapped only affects a case which we've previously said should not
> be supported.

I think it would have no effect at all on setups that don't use
userns, so at least the exposure to potential ABI issues would be
small.

>
> Linus may still disagree - not changing what userspace can do is pretty
> fundamental, but this was purely a missed security fix iiuc.

IIRC I just didn't do it because I didn't want to think about it at
the time, and it didn't look like a *big* security issue.

--Andy