From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Message-ID: <4F4DDE14.8050406@redhat.com> Date: Wed, 29 Feb 2012 09:13:08 +0100 From: Paolo Bonzini MIME-Version: 1.0 To: Linus Torvalds CC: linux-kernel@vger.kernel.org, Jens Axboe , stable@vger.kernel.org, Alan Cox Subject: Re: [PATCH v2] block: avoid false positive warnings on ioctl to partition References: <1329464338-8351-1-git-send-email-pbonzini@redhat.com> <4F4B78CE.4010405@redhat.com> In-Reply-To: Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: Il 29/02/2012 01:14, Linus Torvalds ha scritto: > So I'm still not convinced this is safe, and feel a bit worried about > us possibly silently missing some things. That > > default: > return -ENOIOCTLCMD; > > is what worries me. > > Blocking the ones we *know* about and understand I'm perfectly fine > with. And the SG_IO case looks fine. It's the possibly unknown users > that still worry me. I understand. We do have a good grasp of what's happening. We did get reports for SG_IO, for false positives that would have returned -ENOTTY, and for ioctls that need to be passed. We couldn't expect anything better than this, I think. I checked in the source and all scsi_host-specific ioctls need filtering. Of course we might be missing something really obscure which is rarely used in the wild. But being 100% sure that nothing breaks is impossible, unfortunately, so does it make sense to aim at 100%? And it should be extremely easy to bisect failures. Even with all the differences, it reminds me of the recent change to poll. Paolo