From patchwork Fri Sep 18 05:49:25 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yang Dongsheng X-Patchwork-Id: 7213141 Return-Path: X-Original-To: patchwork-linux-fsdevel@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork1.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.136]) by patchwork1.web.kernel.org (Postfix) with ESMTP id D2B949F32B for ; Fri, 18 Sep 2015 05:56:02 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id E04B220668 for ; Fri, 18 Sep 2015 05:56:01 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id D109220656 for ; Fri, 18 Sep 2015 05:56:00 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752325AbbIRFz7 (ORCPT ); Fri, 18 Sep 2015 01:55:59 -0400 Received: from cn.fujitsu.com ([59.151.112.132]:12028 "EHLO heian.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1751614AbbIRFz6 (ORCPT ); Fri, 18 Sep 2015 01:55:58 -0400 X-IronPort-AV: E=Sophos;i="5.15,520,1432569600"; d="scan'208,223";a="100856002" Received: from bogon (HELO edo.cn.fujitsu.com) ([10.167.33.5]) by heian.cn.fujitsu.com with ESMTP; 18 Sep 2015 13:58:35 +0800 Received: from G08CNEXCHPEKD01.g08.fujitsu.local (localhost.localdomain [127.0.0.1]) by edo.cn.fujitsu.com (8.14.3/8.13.1) with ESMTP id t8I5tRVV000543; Fri, 18 Sep 2015 13:55:27 +0800 Received: from [10.167.226.66] (10.167.226.66) by G08CNEXCHPEKD01.g08.fujitsu.local (10.167.33.89) with Microsoft SMTP Server id 14.3.181.6; Fri, 18 Sep 2015 13:55:43 +0800 Message-ID: <55FBA5E5.1040201@cn.fujitsu.com> Date: Fri, 18 Sep 2015 13:49:25 +0800 From: Dongsheng Yang User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.1.0 MIME-Version: 1.0 To: Jan Kara CC: , , , , Subject: Re: [PATCH v3 11/39] fs: quota: replace opened calling of ->sync_fs with sync_filesystem References: <1442307754-13233-1-git-send-email-yangds.fnst@cn.fujitsu.com> <1442307754-13233-12-git-send-email-yangds.fnst@cn.fujitsu.com> <20150916101402.GE13325@quack.suse.cz> <55FA5D8A.9030108@cn.fujitsu.com> <20150917110547.GB32280@quack.suse.cz> In-Reply-To: <20150917110547.GB32280@quack.suse.cz> X-Originating-IP: [10.167.226.66] Sender: linux-fsdevel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-fsdevel@vger.kernel.org X-Spam-Status: No, score=-6.9 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_HI, T_RP_MATCHES_RCVD, T_TVD_MIME_EPI, UNPARSEABLE_RELAY autolearn=ham version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP On 09/17/2015 07:05 PM, Jan Kara wrote: > On Thu 17-09-15 14:28:26, Dongsheng Yang wrote: >> On 09/16/2015 06:14 PM, Jan Kara wrote: >>> On Tue 15-09-15 17:02:06, Dongsheng Yang wrote: >>>> There are only two places in whole kernel to call ->sync_fs directly. It >>>> will sync fs even in read-only mode. It's not a good idea and some filesystem >>>> would warn out if you are syncing in read-only mode. But sync_filesystem() >>>> does filter this case out. Let's call sync_filesystem() here instead. >>>> >>>> Signed-off-by: Dongsheng Yang >>> >>> So I'd prefer ubifs used hidden system inodes for quota files, set >>> DQUOT_QUOTA_SYS_FILE flag and so this code calling sync_fs() could be >>> completely avoided. >> >> Hmmm, I think it's a good idea to make quota files SYS_FILEs. But how >> about quota-tools? E.g, if quotacheck do some modification >> on quota files, we would not read them without a sync_fs(). >> >> Could you help explain how quota works in this case? > > So tools like quota(1) or setquota(1) work using quotactl so they don't > need access to quota files. When quota files are system files, quotacheck > functionality belongs into the fsck - so fsck.ubifs is responsible for > checking consistency of quota files. How e.g. e2fsck does it is that when > scanning inodes, it computes usage for each user / group, loads limits > information from old quota files and then just creates new quota files with > updated information if there's any difference to the old quota files. About quotacheck, I think just call fsync() in it sounds good to me. Maybe something like the attachment. OKEY, but I found repquota is still using read() to access quota files. Please consider that case: 1.we read quota file and there is a pagecache for it. 2. Then kernel did some modification on quota files. But if the files is SYS_FILES marked, dquot_quota_sync() would not drop the pagecache, then 3. repquota would get an outdated data from pagecache. I am not sure why ext4 works well in this case. There must be something I am missing. Maybe we can introduce a Q_GETBLK for quotactl() to make all quota tools getting informations from ioctl. > > Another advantage of the checking functionality being in fsck is that > fs-specific fsck can gather usage information much more efficiently (and > fsck has to do full fs scan anyway) and there's no need to propagate quota > usage information to userspace using FIQSIZE ioctl() and similar stuff... So, let me try to summary the my opinions about it. Pros: (1). Security. quota files shouldn't be accessible to usespace. (2). Efficiency. No need for quotacheck, just do it in fsck. (3). No need FIOQSIZE any more. Cons: (1). Incompatibility: If I set DQUOT_QUOTA_SYS_FILE currently, there are at least two commands would not work: quotacheck and repquota. Actually that means the whole quota is not usable to user. So, I think the compatibility is important to me. Then what about setting quota files as regular files at first. After all tools (quota tools, quotacheck, repquota, fsck) prepared, setting the DQUOT_QUOTA_SYS_FILE seems better. Yang > > Honza > From 1cf9d072cdd4d17c620834864f79ec5a0765d2bf Mon Sep 17 00:00:00 2001 From: Dongsheng Yang Date: Fri, 18 Sep 2015 08:40:49 -0400 Subject: [PATCH] quotacheck: sync files in quotacheck. After a quotacheck, we need to make kernel to see the modification. Compared with syncing quota files in kernel, we would rather sync it in quotacheck command. Signed-off-by: Dongsheng Yang --- quotacheck.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/quotacheck.c b/quotacheck.c index 6f34cff..e4b184d 100644 --- a/quotacheck.c +++ b/quotacheck.c @@ -768,6 +768,20 @@ rename_new: close(fd); } #endif + + /* Do a syncing for the quota file */ + if ((fd = open(filename, O_RDWR)) < 0) { + errstr(_("Cannot open new quota file %s: %s\n"), filename, strerror(errno)); + free(filename); + return -1; + } + if (fsync(fd)) { + errstr(_("Cannot sync quota file %s: %s\n"), filename, strerror(errno)); + free(filename); + return -1; + } + close(fd); + free(filename); return 0; } -- 1.8.3.1