From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755761AbZCWJ2q (ORCPT ); Mon, 23 Mar 2009 05:28:46 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752332AbZCWJ2g (ORCPT ); Mon, 23 Mar 2009 05:28:36 -0400 Received: from zeniv.linux.org.uk ([195.92.253.2]:36979 "EHLO ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752173AbZCWJ2f (ORCPT ); Mon, 23 Mar 2009 05:28:35 -0400 Date: Mon, 23 Mar 2009 09:28:30 +0000 From: Al Viro To: Eric Paris Cc: linux-kernel@vger.kernel.org, hch@infradead.org, alan@lxorguk.ukuu.org.uk, sfr@canb.auug.org.au, john@johnmccutchan.com, rlove@rlove.org, akpm@linux-foundation.org Subject: Re: [PATCH 04/13] fsnotify: add in inode fsnotify markings Message-ID: <20090323092830.GN28946@ZenIV.linux.org.uk> References: <20090319180456.17449.76663.stgit@paris.rdu.redhat.com> <20090319180514.17449.34794.stgit@paris.rdu.redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20090319180514.17449.34794.stgit@paris.rdu.redhat.com> User-Agent: Mutt/1.5.17 (2007-11-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Mar 19, 2009 at 02:05:14PM -0400, Eric Paris wrote: > This patch creates in inode fsnotify markings. dnotify will make use of in > inode markings to mark which inodes it wishes to send events for. fanotify > will use this to mark which inodes it does not wish to send events for. Hrm... What happens if fsnotify_clear_marks_by_inode() is called, gets the only mark we had for a group, removes it from all lists, drops all locks and races with the final fsnotify_put_group()? Said put_group sees nothing whatsoever on the per-group mark list (the only mark we used to have there is already evicted) and the group is cheerfully freed. Then fsnotify_destroy_mark_by_entry() calls group->ops->freeing_mark(entry, group) with group already freed... Gets really interesting in case of inotify - there ->freeing_mark() leads to blocking allocations and really touches group in non-trivial ways. Note that for inotify ->freeing_mark() will happily add notification to group's queue and that's another race of the same nature - even if group itself isn't yet freed, that final put_group() might bloody well get past flushing the queue. I really don't like the idea of having marks contributing to group refcount - that would have solved this one, but we'd get potential leaks from hell ;-/ Maybe a separate counter controlling only actual freeing of group?