From: vcaputo@gnugeneration.com Subject: OverlayFS limitation Date: Tue, 14 Apr 2015 15:45:21 -0500 Message-ID: <20150414204521.GB8110@shells.gnugeneration.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: vito.caputo@coreos.com To: linux-fsdevel@vger.kernel.org Return-path: Received: from shells.gnugeneration.com ([66.240.222.126]:50841 "HELO shells.gnugeneration.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1755056AbbDNUwD (ORCPT ); Tue, 14 Apr 2015 16:52:03 -0400 Content-Disposition: inline Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Hello fsdevel, I've been poking around overlayfs recently and it appears one of the major remaining barriers to general usage as a posix-ish filesystem is reader file handles accessing stale (lower) instances of copy_up'd files. Can't we do something to invalidate the struct address_space * of the lower instance's inode referenced by the open files when the copy_up occurs? I don't see any formal infrastructure supporting this currently, other than maybe the scaffolding for truncate, but it seems tractable, if perhaps a bit invasive, to make happen. Anyone already been down this rabbit hole? Cheers, Vito Caputo