From mboxrd@z Thu Jan 1 00:00:00 1970 From: Zach Brown Subject: [PATCH RFC v2 0/2] O_NOCMTIME protected by generic mount option Date: Fri, 15 May 2015 14:23:46 -0700 Message-ID: <1431725028-24071-1-git-send-email-zab@zabbo.net> To: Sage Weil , linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, linux-api@vger.kernel.org Return-path: Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org Here's a current draft of what is now the O_NOCMTIME series. It implements the frequent suggestion to gate unprivileged O_NOCMTIME use with a mount option. This method has the advantage of being entirely runtime. There's no persistence that'd require updating all the tools that deal with each file system's format. It's also requested by writers as they open. Writes to the file that know nothing of O_NOCMTIME will behave as usual. Another suggested method is to use inode attributes: require root to set +nocmtime on a dir and inherit it down subdirs to new files. This nicely solves the unprivilieged use problem without having to fiddle with mount options but it requires touching all systems that support it and would prevent cmtime updates on all writes to the inode. I have a patch series that starts on this but haven't taken it very far. Sage is working on spinning up some hardware to test the various dirty inode avoidance methods at load and should have numbers soon. That'll tell us if lazytime isn't good enough and if any of this is worth the trouble. - z