From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from zeniv.linux.org.uk ([195.92.253.2]:42786 "EHLO ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751274AbeBWEIP (ORCPT ); Thu, 22 Feb 2018 23:08:15 -0500 Date: Fri, 23 Feb 2018 04:08:14 +0000 From: Al Viro To: John Ogness Cc: linux-fsdevel@vger.kernel.org, Linus Torvalds , Christoph Hellwig , Thomas Gleixner , Peter Zijlstra , Sebastian Andrzej Siewior , linux-kernel@vger.kernel.org Subject: Re: [PATCH v2 6/6] fs/dcache: Avoid remaining try_lock loop in shrink_dentry_list() Message-ID: <20180223040814.GA30522@ZenIV.linux.org.uk> References: <20180222235025.28662-1-john.ogness@linutronix.de> <20180222235025.28662-7-john.ogness@linutronix.de> <20180223035814.GZ30522@ZenIV.linux.org.uk> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180223035814.GZ30522@ZenIV.linux.org.uk> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Fri, Feb 23, 2018 at 03:58:14AM +0000, Al Viro wrote: > On Fri, Feb 23, 2018 at 12:50:25AM +0100, John Ogness wrote: > > > Avoid the trylock loop by using dentry_kill(). When killing dentries > > from the dispose list, it is very similar to killing a dentry in > > dput(). The difference is that dput() expects to be the last user of > > the dentry (refcount=1) and will deref whereas shrink_dentry_list() > > expects there to be no user (refcount=0). In order to handle both > > situations with the same code, move the deref code from dentry_kill() > > into a new wrapper function dentry_put_kill(), which can be used > > by previous dentry_kill() users. Then shrink_dentry_list() can use > > the dentry_kill() to cleanup the dispose list. > > > > This also has the benefit that the locking order is now the same. > > First the inode is locked, then the parent. > > Current code moves the sucker to the end of list in that case; I'm not > at all sure that what you are doing will improve the situation at all... > > You *still* have a trylock loop there - only it keeps banging at the > same dentry instead of going through the rest first... Actually, it's even worse - _here_ you are dealing with something that really can change inode under you. This is one and only case where we are kicking out a zero-refcount dentry without having already held ->i_lock. At the very least, it's bloody different from regular dentry_kill(). In this case, dentry itself is protected from freeing by being on the shrink list - that's what makes __dentry_kill() to leave the sucker allocated. We are not holding references, it is hashed and anybody could come, pick it, d_delete() it, etc.