From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 8676E36F415 for ; Sat, 9 May 2026 08:51:19 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=10.30.226.201 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1778316679; cv=none; b=Q8XPE7efIccPj8JSU5VBBCI4nGx0eAH0ok4wOkr0goYg+cUAAvAM7CplIrPgYTvz2U+7u1+5kjqWv5PFKv8MruxQVCEcyP2w3SQmA2URhaX4ubDtiCYQ+lItuxP72R4Q8/u1niQWrnkRefkIDxftovUwdn1/qINOd+79Uvmtlws= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1778316679; c=relaxed/simple; bh=eHqFRbXmmGGDZMkdm77B566cULy87KKBa/U4vdrLq1o=; h=Message-ID:Date:MIME-Version:Cc:Subject:To:References:From: In-Reply-To:Content-Type; b=GyR/cOL59Tu1HN/56HhINB5+xKr2sYXIknopnBnOIqiaSekM/iB247C2Nf5dj2kSWiOgx9dvv+3T0l6JtkaWwehHrtjlx/JP4/gr4CWe0i61gmvni2zDHarBDHektXxVJrYgKskX4ys2t1O1w8v8Mmk7gRqjBwH9GiL9Rab3lrc= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=e9FaN3t5; arc=none smtp.client-ip=10.30.226.201 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="e9FaN3t5" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 327EDC2BCB2; Sat, 9 May 2026 08:51:16 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1778316679; bh=eHqFRbXmmGGDZMkdm77B566cULy87KKBa/U4vdrLq1o=; h=Date:Cc:Subject:To:References:From:In-Reply-To:From; b=e9FaN3t5FRM97GqC9xQnlTkQUqxqfHI6EBgH2Zwuk8DvzkXdMOkmELzMaRP+eq4mj JKxj7w1IbuNoG3nQZhcD1cdwBo0Rls3Zs/nxJhGkRCSbyYIsfpHC5rT5uXH/7t+LAE MPAnZgeTxMLSrAlf7tal7e08W8CaK8VVe1Gu1eTMxZqdRp/qu5B7Q5QmZ4G6fsbiEc 0utMEXpz2Zhc06Kp4abT+3lNT44UVOnLNe1pAIernzoefbn8p9rWR5Y6RqtYwarXdJ 7G9oJpPiAbl7ipSIC5rapoKUCJjD/pIBUPC9ys8Sh9OlgsFkAJb0qF4N59Z6umm2nh OqtE9PmjHrQQQ== Message-ID: Date: Sat, 9 May 2026 16:51:14 +0800 Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Cc: chao@kernel.org, jaegeuk@kernel.org, quic_stummala@quicinc.com, linux-f2fs-devel@lists.sourceforge.net, linux-kernel@vger.kernel.org, baijiaju1990@gmail.com Subject: Re: [PATCH] f2fs: start discard thread after mount recovery To: Cen Zhang References: <20260503045820.973718-1-zzzccc427@gmail.com> <2ba0b3ad-615e-43c8-a165-af631607a4c3@kernel.org> Content-Language: en-US From: Chao Yu In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit On 5/3/26 18:16, Cen Zhang wrote: > Dear Chao Yu > > Thanks for taking a look, and sorry for the confusion. > >> >> On 5/3/26 12:58, Cen Zhang wrote: >>> The discard command control is built while the segment manager is >>> constructed, before the node manager is built and before mount recovery >>> has completed. Starting the discard thread from that constructor lets the >>> background thread run while f2fs_fill_super() is still publishing and >>> initializing mount-time state. >>> >>> After commit d6d2b491a82e ("f2fs: allow to change discard policy based >>> on cached discard cmds"), issue_discard_thread() may consult node-manager >>> memory thresholds through f2fs_available_free_memory(). It can therefore >>> observe sbi->nm_info while f2fs_build_node_manager() is publishing and >>> initializing it. The same early-start window also lets the thread test >>> the superblock read-only state while recovery paths still make temporary >>> SB_RDONLY transitions. >> >> Not sure I understood you correctly, do you mean this? >> > No, I did not mean the old UAF-style failure path. That path should > already be covered by the existing cleanup ordering: > > free_nm: > f2fs_stop_discard_thread(sbi); > f2fs_destroy_node_manager(sbi); > Cen, can you please add all below details into v3? Thanks, > What I was trying to describe is an earlier window during mount > initialization, when the discard thread is actually started: > > - f2fs_fill_super() > - f2fs_build_segment_manager() > - create_discard_cmd_control() > - f2fs_start_discard_thread() > - f2fs_build_node_manager() > - sbi->nm_info = f2fs_kzalloc(...) > - init_node_manager() initializes fields such as ram_thresh, counters, > locks/lists/bitmaps, etc. > > At this point issue_discard_thread() may run concurrently and call > f2fs_available_free_memory(). That helper reads NM_I(sbi), returns true > only if it is NULL, and otherwise uses node-manager fields such as > nm_i->ram_thresh. So my concern is that the thread may observe a newly > published but still being initialized node manager, rather than a freed > one. > > The data race report maps to the following paths in v7.0.3: > > issue_discard_thread() fs/f2fs/segment.c:1921 > f2fs_available_free_memory() fs/f2fs/node.c:50 > NM_I() fs/f2fs/f2fs.h:2228 > > racing with mount-time node-manager initialization: > > f2fs_fill_super() fs/f2fs/super.c:5151 > f2fs_build_node_manager() fs/f2fs/node.c:3420 > > There is also a similar early-start window for the SB_RDONLY check in > issue_discard_thread(), since mount recovery can still make temporary > superblock flag transitions before f2fs_fill_super() reaches the stable > mounted state. The second report maps to: > > issue_discard_thread() fs/f2fs/segment.c:1935 > f2fs_readonly() fs/f2fs/f2fs.h:3665 > > racing with mount recovery restoring the superblock flags: > > f2fs_fill_super() fs/f2fs/super.c:5267 > f2fs_recover_fsync_data() fs/f2fs/recovery.c:953 > > So the patch is intended to fix this lifecycle ordering: keep the discard > command control available early, but defer starting the background > discard thread until mount recovery and node-manager initialization have > completed. > > Please let me know if I missed anything here. > > Best regards, > Cen