Re: [PATCH net-next] modules: allow modprobe load regular elf binaries

From: Andy Lutomirski <luto@amacapital.net>
To: David Miller <davem@davemloft.net>
Cc: luto@kernel.org, alexei.starovoitov@gmail.com,
	keescook@chromium.org, ast@kernel.org, tixxdz@gmail.com,
	viro@zeniv.linux.org.uk, daniel@iogearbox.net,
	torvalds@linux-foundation.org, gregkh@linuxfoundation.org,
	mcgrof@kernel.org, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org, kernel-team@fb.com,
	linux-api@vger.kernel.org
Subject: Re: [PATCH net-next] modules: allow modprobe load regular elf binaries
Date: Thu, 8 Mar 2018 19:10:54 -0800	[thread overview]
Message-ID: <3BC1EAA3-D926-4758-901D-A860718B846A@amacapital.net> (raw)
In-Reply-To: <20180308.213153.2003279953084099668.davem@davemloft.net>

> On Mar 8, 2018, at 6:31 PM, David Miller <davem@davemloft.net> wrote:
> 
> From: Andy Lutomirski <luto@kernel.org>
> Date: Fri, 9 Mar 2018 02:12:24 +0000
> 
>> First, compile your user code and emit a staitc binary.  Use objdump
>> fiddling or a trivial .S file to make that static binary into a
>> variable.  Then write a tiny shim module like this:
>> 
>> extern unsigned char __begin_user_code[], __end_user_code[];
>> 
>> int __init init_shim_module(void)
>> {
>>  return call_umh_blob(__begin_user_code, __end_user_code - __begin_user_code);
>> }
>> 
>> By itself, this is clearly a worse solution than yours, but it has two
>> benefits, one small and two big.  The small benefit is that it is
>> completely invisible to userspace: the .ko file is a bona fide module.
> 
> Anything you try to do which makes these binaries "special" is a huge
> negative.

I don’t know what you mean.  Alexei’s approach introduces a whole new kind of special module.  Mine doesn’t. 

> 
>> The big benefits are:
> 
> I don't see those things as benefits at all, and Alexei's scheme can
> easily be made to work in your benefit #1 case too.
> 

How?  I think you’ll find that a non-modular implementation of a bundled ELF binary looks a *lot* like my call_umh_blob().

> It's a user binary.  It's shipped with the kernel and it's signed.
> 
> If we can't trust that, we can't trust much else.

I’m not making any arguments about security at all. I’m talking about functionality. 

If we apply Alexei’s patch as is, then I think we’ll have a situation where ET_EXEC modules are only useful if they can do their jobs without any filesystem access at all.  This is fine for networking, where netlink sockets are used, but I think it’s not so great for other use cases. If we ever try to stick a usb driver into userspace, we’re going to want to instantiate the user task once per device, passed as stdin or similar, and Alexei’s code will make that very awkward.