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;
Subject: Re: [PATCH RFC 2/3] io_uring: add IOURING_REGISTER_RESTRICTIONS
 opcode
To:     Stefano Garzarella <sgarzare@redhat.com>
Cc:     Sargun Dhillon <sargun@sargun.me>,
        Kees Cook <keescook@chromium.org>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        Kernel Hardening <kernel-hardening@lists.openwall.com>,
        Jann Horn <jannh@google.com>, Aleksa Sarai <asarai@suse.de>,
        Christian Brauner <christian.brauner@ubuntu.com>,
        Stefan Hajnoczi <stefanha@redhat.com>,
        io-uring@vger.kernel.org, Alexander Viro <viro@zeniv.linux.org.uk>,
        Jeff Moyer <jmoyer@redhat.com>
References: <20200710141945.129329-1-sgarzare@redhat.com>
 <20200710141945.129329-3-sgarzare@redhat.com>
From:   Jens Axboe <axboe@kernel.dk>
Message-ID: <f39fe84d-1353-1066-c7fc-770054f7129e@kernel.dk>
Date:   Fri, 10 Jul 2020 11:52:48 -0600
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <20200710141945.129329-3-sgarzare@redhat.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 7/10/20 8:19 AM, Stefano Garzarella wrote:
> The new io_uring_register(2) IOURING_REGISTER_RESTRICTIONS opcode
> permanently installs a feature whitelist on an io_ring_ctx.
> The io_ring_ctx can then be passed to untrusted code with the
> knowledge that only operations present in the whitelist can be
> executed.
> 
> The whitelist approach ensures that new features added to io_uring
> do not accidentally become available when an existing application
> is launched on a newer kernel version.

Keeping with the trend of the times, you should probably use 'allowlist'
here instead of 'whitelist'.
> 
> Currently is it possible to restrict sqe opcodes and register
> opcodes. It is also possible to allow only fixed files.
> 
> IOURING_REGISTER_RESTRICTIONS can only be made once. Afterwards
> it is not possible to change restrictions anymore.
> This prevents untrusted code from removing restrictions.

A few comments below.

> @@ -337,6 +344,7 @@ struct io_ring_ctx {
>  	struct llist_head		file_put_llist;
>  
>  	struct work_struct		exit_work;
> +	struct io_restriction		restrictions;
>  };
>  
>  /*

Since very few will use this feature, was going to suggest that we make
it dynamically allocated. But it's just 32 bytes, currently, so probably
not worth the effort...

> @@ -5491,6 +5499,11 @@ static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
>  	if (unlikely(!fixed && io_async_submit(req->ctx)))
>  		return -EBADF;
>  
> +	if (unlikely(!fixed && req->ctx->restrictions.enabled &&
> +		     test_bit(IORING_RESTRICTION_FIXED_FILES_ONLY,
> +			      req->ctx->restrictions.restriction_op)))
> +		return -EACCES;
> +
>  	return io_file_get(state, req, fd, &req->file, fixed);
>  }

This one hurts, though. I don't want any extra overhead from the
feature, and you're digging deep in ctx here to figure out of we need to
check.

Generally, all the checking needs to be out-of-line, and it needs to
base the decision on whether to check something or not on a cache hot
piece of data. So I'd suggest to turn all of these into some flag.
ctx->flags generally mirrors setup flags, so probably just add a:

	unsigned int restrictions : 1;

after eventfd_async : 1 in io_ring_ctx. That's free, plenty of room
there and that cacheline is already pulled in for reading.


-- 
Jens Axboe