From mboxrd@z Thu Jan 1 00:00:00 1970 From: Toshiyuki Okajima Subject: Re: [PATCH -v3] vfs: add releasepages hooks to block devices which can be used by file systems Date: Thu, 18 Dec 2008 14:15:25 +0900 Message-ID: <4949DC6D.3050908@jp.fujitsu.com> References: <20081212062148.GJ10890@mit.edu> <1229104375-11567-1-git-send-email-tytso@mit.edu> <20081217153940.GA6495@atrey.karlin.mff.cuni.cz> Reply-To: toshi.okajima@jp.fujitsu.com Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: "Theodore Ts'o" , Ext4 Developers List , linux-fsdevel@vger.kernel.org To: Jan Kara Return-path: Received: from fgwmail7.fujitsu.co.jp ([192.51.44.37]:33516 "EHLO fgwmail7.fujitsu.co.jp" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750757AbYLRFPi (ORCPT ); Thu, 18 Dec 2008 00:15:38 -0500 In-Reply-To: <20081217153940.GA6495@atrey.karlin.mff.cuni.cz> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Hi, > Hello, > > > > From: Toshiyuki Okajima > > > > > > Implement blkdev_releasepage() to release the buffer_heads and page > > > after we release private data which belongs to a client of the block > > > device, such as a filesystem. > > > > > > blkdev_releasepage() call the client's releasepage() which is > > > registered by blkdev_register_client_releasepage() to release its > > > private data. > Yes, this is IMO the right fix. I'm just wondering about the fact that we > can't block in the client_releasepage(). That seems to be caused by the fact > that we need to be protected against client_releasepage() callback changes > which essentially means umount, right? I'm not saying I have a better solution > but introducing such limitation seems stupid just because of umount... > > Honza > Difference between v2 and v3 in blkdev_releasepage: < ret = (*ei->client_releasepage)(ei->client, page, wait); < else -- > /* > * Since we are holding a spinlock (ei->client_lock), > * make sure the client_releasepage function > * understands that it must not block. > */ > ret = (*ei->client_releasepage)(ei->client, page, > wait & ~__GFP_WAIT); > else Ask for clarification. Which of the following do you mean: 1) If using a spinlock in client_releasepage() is only for mount/umount, this implementation is not wise. 2) There is the fact that a spinlock is necessary for blkdev_releasepage(). This fact prevents us from making various implementations of client_releasepage(). (Without a spinlock, we can implement a client_releasepage() which can release the buffers with a sleep. As a result, it may enable more buffers release than before.) There is the fact that a filesystem can be mounted on several places, and the lock mechanism is absolutely necessary for this fact. I also think we are sad that we cannot implement various implementations for client_releasepage(). But now I cannot imagine what to do for a client_releasepage() which can sleep, too... Regards, Toshiyuki Okajima