From: Theodore Ts'o Subject: Re: [PATCH 2/4] ext4: fix deadlock between inline_data and ext4_expand_extra_isize_ea() Date: Tue, 24 Jan 2017 20:32:48 -0500 Message-ID: <20170125013248.lmxkc4ruh66o4p7b@thunk.org> References: <20170112034938.5934-1-tytso@mit.edu> <20170112034938.5934-4-tytso@mit.edu> <20170120135317.GC10446@quack2.suse.cz> <20170122222527.ylc26specpuvydcx@thunk.org> <20170124122722.GE20153@quack2.suse.cz> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Ext4 Developers List , linux@sciencehorizons.net, stable@vger.kernel.org, #@thunk.org To: Jan Kara Return-path: Content-Disposition: inline In-Reply-To: <20170124122722.GE20153@quack2.suse.cz> Sender: stable-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org On Tue, Jan 24, 2017 at 01:27:22PM +0100, Jan Kara wrote: > I see, thanks for explanation. Well seeing all these problems with > ext4_expand_extra_isize() wouldn't we be better off by not calling it from > ext4_mark_inode_dirty() but rather explicitely from several well-defined > places? Because this implicit calling looks like it causes us too much > trouble. Yeah, I suppose that might be a better way to go. We could only do it on a file open, perhaps. It might not be as important it is to expand the extra_isize on, say, a chmod, for example. It's certainly worth looking to see whether it would simplify things to go that way. - Ted