Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sun, 17 May 2020 14:30:57 -0700
From:   Kees Cook <keescook@chromium.org>
To:     Tycho Andersen <tycho@tycho.ws>
Cc:     Christian Brauner <christian.brauner@ubuntu.com>,
        Aleksa Sarai <asarai@suse.de>, linux-api@vger.kernel.org,
        containers@lists.linux-foundation.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] seccomp: Add group_leader pid to seccomp_notif
Message-ID: <202005171428.68F30AA0@keescook>
References: <20200515234005.32370-1-sargun@sargun.me>
 <202005162344.74A02C2D@keescook>
 <20200517104701.bbn2d2rqaplwchdw@wittgenstein>
 <20200517112156.cphs2h33hx2wfcs4@yavin.dot.cyphar.com>
 <20200517142316.GA1996744@cisco>
 <20200517143311.fmxaf3pnopuaezl4@wittgenstein>
 <20200517144603.GD1996744@cisco>
 <20200517150215.GE1996744@cisco>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200517150215.GE1996744@cisco>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, May 17, 2020 at 09:02:15AM -0600, Tycho Andersen wrote:
> On Sun, May 17, 2020 at 08:46:03AM -0600, Tycho Andersen wrote:
> > On Sun, May 17, 2020 at 04:33:11PM +0200, Christian Brauner wrote:
> > > struct seccomp_notif2 {
> > > 	__u32 notif_size;
> > > 	__u64 id;
> > > 	__u32 pid;
> > > 	__u32 flags;
> > > 	struct seccomp_data data;
> > > 	__u32 data_size;
> > > };
> > 
> > I guess you need to put data_size before data, otherwise old userspace
> > with a smaller struct seccomp_data will look in the wrong place.
> > 
> > But yes, that'll work if you put two sizes in, which is probably
> > reasonable since we're talking about two structs.
> 
> Well, no, it doesn't either. Suppose we add a new field first to
> struct seccomp_notif2:
> 
> struct seccomp_notif2 {
>     __u32 notif_size;
>     __u64 id;
>     __u32 pid;
>     __u32 flags;
>     struct seccomp_data data;
>     __u32 data_size;
>     __u32 new_field;
> };
> 
> And next we add a new field to struct secccomp_data. When a userspace
> compiled with just the new seccomp_notif2 field does:
> 
> seccomp_notif2.new_field = ...;
> 
> the compiler will put it in the wrong place for the kernel with the
> new seccomp_data field too.
> 
> Sort of feels like we should do:
> 
> struct seccomp_notif2 {
>     struct seccomp_notif *notif;
>     struct seccomp_data *data;
> };

I'm going read this thread more carefully tomorrow, but I just wanted to
mention that I'd *like* to extend seccomp_data for doing deep argument
inspection of the new syscalls. I think it's the least bad of many
designs, and I'll write that up in more detail. (I would *really* like
to avoid extending seccomp's BPF language, and instead allow probing
into the struct copied from userspace, etc.)

Anyway, it's very related to this, so, yeah, probably we need a v2 of the
notif API, but I'll try to get all the ideas here collected in one place.

-- 
Kees Cook