From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: ebiederm@xmission.com (Eric W. Biederman) To: Jeff Layton Cc: viro@zeniv.linux.org.uk, berrange@redhat.com, linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, Oleg Nesterov References: <20180914105310.6454-1-jlayton@kernel.org> Date: Sun, 16 Sep 2018 19:38:55 +0200 In-Reply-To: <20180914105310.6454-1-jlayton@kernel.org> (Jeff Layton's message of "Fri, 14 Sep 2018 06:53:07 -0400") Message-ID: <87a7ohs5ow.fsf@xmission.com> MIME-Version: 1.0 Content-Type: text/plain Subject: [RFC][PATCH 0/3] exec: Moving unshare_files_struct Sender: linux-kernel-owner@vger.kernel.org List-ID: Paired with Oleg's patch to reduce the number of callers of get_files_struct it looks like we can simplify the basic idea of moving unshare_files in exec by quite a bit so that in net we have fewer lines of code. The big simplification from Jeff's verion is that we take advantage of calling unshare_files past the point of no return. Which removes the need for cleanup, and restoring ->files. Which removes the need for blocking clone and unshare. Oleg's patch to remove get_files_struct from proc means we don't need two counts in files_struct. Which leaves us with the question of what are the races in fs/exec.c with respect to accessing files. Semantically I don't think we care but we do need to be certain the implementation of exec is still robust. These patches are still rough and ready and only compile tested but I believe they demonstrate what is possible. Eric W. Biederman (3): exec: Move unshare_files down to avoid locks being dropped on exec. exec: Simplify unshare_files exec: Remove reset_files_struct fs/coredump.c | 5 +---- fs/exec.c | 16 +++++----------- fs/file.c | 12 ------------ include/linux/fdtable.h | 3 +-- kernel/fork.c | 12 ++++++------ 5 files changed, 13 insertions(+), 35 deletions(-) Eric