Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752018AbaGUBmG (ORCPT ); Sun, 20 Jul 2014 21:42:06 -0400 Received: from youngberry.canonical.com ([91.189.89.112]:59612 "EHLO youngberry.canonical.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751598AbaGUBmE (ORCPT ); Sun, 20 Jul 2014 21:42:04 -0400 MIME-Version: 1.0 In-Reply-To: References: <1405373897-31671-1-git-send-email-keescook@chromium.org> <1405373897-31671-7-git-send-email-keescook@chromium.org> Date: Mon, 21 Jul 2014 09:42:00 +0800 Message-ID: Subject: Re: [PATCH 6/7] firmware_class: add "fd" input file From: Ming Lei To: Kees Cook Cc: Linux Kernel Mailing List , "Luis R. Rodriguez" , Greg Kroah-Hartman , James Morris , David Howells , "linux-doc@vger.kernel.org" , linux-security-module , linux-firmware@kernel.org, linux-wireless Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, Jul 21, 2014 at 1:43 AM, Kees Cook wrote: > On Sat, Jul 19, 2014 at 8:04 PM, Ming Lei wrote: >> On Tue, Jul 15, 2014 at 5:38 AM, Kees Cook wrote: >>> As an alternative to loading bytes from the "data" blob when reading >>> firmware, let kernel read from an fd, so that the LSM can reason about >>> the origin of firmware contents during userspace on-demand loading. >> >> From user space view, maybe it is better to keep previous usage and just >> check if loading is from 'data' blob or fd in 'echo 0 > loading' of >> firmware_loading_store(), then the 'fd' usage becomes very similar with >> before. > > I don't think this is a good idea because otherwise there isn't a good > way to have an "atomic" check of the firmware contents. What does it Could you share why 'atomic' check is necessary? As we know, it isn't real atomic, :-). > means to write to "fd" several times, then write "data" a little, > before writing "loading", etc? I originally wrote the patch series That depends how firmware loader supports these cases, and won't be difficult to handle them. For non-fd userspace interface, it is very flexible to be capable of supporting to load firmware data from multiple images, or in flight. With single 'fd' interface, it won't be possible any more. > requiring the "loading" piece, and it ended up being very complicated > due to needing to switch the memory buffer logic back and forth. > Everything is much much cleaner if "fd" is single-shot, and not part > of the loading/data interface semantics. You might not avoid the 'loading' piece completely, how does the userspace handle non-exist firmware image? The reasonable way is to abort the loading from userspace via 'echo -1 > loading' since userspace already sees that, and you may choose to let kernel side handle that, but your current patch doesn't support it yet. Thanks, -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/