From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mail-vk0-f46.google.com ([209.85.213.46]:40510 "EHLO mail-vk0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752979AbeDBQOy (ORCPT ); Mon, 2 Apr 2018 12:14:54 -0400 MIME-Version: 1.0 In-Reply-To: References: <20180326225921.10096-1-fdmanana@kernel.org> <20180328021753.GE30836@localhost.localdomain> <20180328103358.GG30836@localhost.localdomain> From: Jayashree Mohan Date: Mon, 2 Apr 2018 11:14:53 -0500 Message-ID: Subject: Re: [PATCH] fstests: test btrfs fsync after hole punching with no-holes mode Content-Type: text/plain; charset="UTF-8" Sender: fstests-owner@vger.kernel.org To: Filipe Manana Cc: Eryu Guan , fstests , linux-btrfs , Filipe Manana , Vijaychidambaram Velayudhan Pillai List-ID: On Mon, Apr 2, 2018 at 9:24 AM, Filipe Manana wrote: > On Thu, Mar 29, 2018 at 7:55 PM, Jayashree Mohan > wrote: >> Hi Filipe, >> >> I tried running the xfstest above on kernel 4.15.0 and I haven't been >> able to hit the bug. The xfstest passes clean for me. I compared the >> btrfs-debug-tree in my case with the one attached by Eryu, and I see I >> do not have any xattr as he does. However, for every run of the >> xfstest, the extent tree info that I get from btrfs-debug-tree has >> varying number of items in the first leaf node. (I have attached two >> dump files for your reference.) >> >> I also tried changing the offset at which fpunch is performed to match >> the offset of the last extent in the first leaf of the extent tree - >> however the md5 before and after crash was the same. >> >> Could you give more details on this? > > You are getting extents smaller then 256Kb, because writeback is being > triggered by the kernel (likely due to memory pressure). > The second version of the test uses direct IO instead to avoid that. > Thanks. Thanks for the clarification. I am able to reproduce the bug with the new version of the test that uses direct writes.