From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-alma10-1.taild15c8.ts.net [100.103.45.18]) (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 3AB123C8C7D for ; Mon, 12 Oct 2026 09:25:07 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=100.103.45.18 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1791797109; cv=none; b=O19tK4Mfcqv4ci3ZY7edOhD/YZCczevpgzKamGIuecf/CD2ohw0SCHThDkRolcSqGsdq1KxaDaeAAte5/NrQD+PhWG8G/9QM903GMVtRpWnv6eeAML8wrKUg4QbGgd05ywlECRrY6WzLbRQcHz8UTuX7VM0Xepo6HRxjXAmQ0F8= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1791797109; c=relaxed/simple; bh=DVCB0TluVLSYS+xBk2cX1xzrIC9QXbmVKD1YxRQ9G68=; h=Message-ID:Date:MIME-Version:Cc:Subject:To:References:From: In-Reply-To:Content-Type; b=cpFhqTjiCspQNq04E4FTOVx6odclJsSyXUauBnzIJO3Td2I1z3HcV9ltdTSIk+25al0Y9knipzS05m5H2CupkOHN5UFYD1Wy7esE/cu5VcfmJKHMyFFw7l3gvByKCc5IgE3/K30ceKb0YSwvQXw8N6sM0OeWRbMr96y6B5oykjo= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=FyiS94XE; arc=none smtp.client-ip=100.103.45.18 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="FyiS94XE" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 350C31F0089B; Mon, 12 Oct 2026 09:25:05 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel.org; s=k20260515; t=1791797107; bh=0PKUd2tYPRTuC5yA1fFc//1j3Cjcv14nzwJE45Zv7sM=; h=Date:Cc:Subject:To:References:From:In-Reply-To; b=FyiS94XEshUNU+NDRYiayutlJ5/K4I53pGGTAFoseU7K1NZNLqozVRp7cXZhGTjqm RJkzGG+/bJs4l7SH+UGNGyojbH9PH+vGr0O+OeH00FZwmQrOevGlavMYcoeCKH7+yD hJLSU74qhLV3tawIQ9XhHbENPkj2XtJ7fGgbPKncAn4jipqKDBiWW6NL+Z4bCzSqgV EBn69wMfX/x8mLZ4MgbEDYnCmMKBjhpDF/uJu+RXmFjV9SKbvyAsUqEk7y1jr2f+L/ I4mmxLgZHQyNq7luH3B3PEMHtFWYmd1ygzUYTVy0t0guPTvJJyqBWOJq0hm/YObEeK Vu2KQiJlnfgKQ== Message-ID: <6cb0ccd5-bd2e-49e9-9f87-83b4302d110b@kernel.org> Date: Mon, 12 Oct 2026 17:25:03 +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, linux-kernel@vger.kernel.org, linux-f2fs-devel@lists.sourceforge.net, kernel-team@android.com, Daeho Jeong Subject: Re: [f2fs-dev] [PATCH] f2fs: cache: wake up the writeback thread while checkpoint is disabled To: Daeho Jeong References: <20261007193131.1785199-1-daeho43@gmail.com> Content-Language: en-US From: Chao Yu In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit On 10/11/26 23:34, Daeho Jeong wrote: > On Sun, Oct 11, 2026 at 1:22 AM Chao Yu wrote: >> >> On 10/8/26 03:31, Daeho Jeong wrote: >>> From: Daeho Jeong >>> >>> Dirty node and meta caches are written back by the cache writeback >>> thread and by checkpoint. The thread wakes up every cache_wb_interval >>> (5 seconds by default) and writes at most 512 node caches and one >>> contiguous run of meta caches. Checkpoint, which f2fs_balance_fs_bg() >>> triggers when there are too many dirty caches, writes the rest. >>> >>> While checkpoint is disabled, checkpoint does not run: f2fs_sync_fs() >>> returns early and f2fs_balance_fs() returns before doing anything. So >>> dirty node caches pile up without a limit. They are kmalloc'd, and the >>> shrinker skips dirty entries, so this memory cannot be reclaimed. When >>> node and meta blocks were in the page cache, the dirty pages were >>> counted as dirty page cache, and the flusher threads wrote them back in >>> the background. >>> >>> A checkpoint=disable test that remounts the filesystem with >>> checkpoint=disable and runs fsstress -p 32 for 300 seconds left about >>> 160K dirty node caches (2.7GB of kmalloc-16k) with 16KB blocks and >>> about 226K with 4KB blocks in QEMU, and drop_caches freed none of them. >>> >>> While checkpoint is disabled, wake up the thread from >>> f2fs_mark_cache_dirty() once there are as many dirty node or meta >>> caches as f2fs_write_node_caches() and f2fs_write_meta_caches() collect >>> before writing, and let the thread go on without waiting while that is >>> still true and each round writes some of them. Each round writes the >>> same number of caches as before. Split nr_caches_to_collect() out of >>> nr_caches_to_skip() to share these numbers without the dirty_exceeded >>> check. >>> >>> With this, the same test keeps about 4K dirty node caches with both >>> 4KB and 16KB blocks. Nothing changes while checkpoint is enabled. >> >> I've sent a patch to wake up f2fs_writeback to flush dirty data according >> to configurable sysfs value, I guess it can fix your checkpoint disable >> case, please take a look. >> >> [PATCH 3/5] f2fs: cache: wake up f2fs_writeback when exceeding threshold >> >> Thanks, > > Hi Chao, > > Thanks, I think it covers my case. Hi Daeho, > > Two things in v2: > > 1. The wait condition stays true when the thread can't reduce the dirty > count, so it loops without sleeping: on f2fs_cp_error(), readonly, > freezing(current) or a failed sb_start_write_trylock() (dirty caches > stay after shutdown/freeze with checkpoint=disable, and during > suspend the thread may never freeze), or when the threshold is below > nr_caches_to_skip() so the write functions write nothing. How about > sleeping for the interval when a round doesn't reduce the dirty > count? Yeah, I see, I figured out a throttle field to avoid unnecessary wakeup in v3. > > 2. With both thresholds 0, dirty caches grow again under > checkpoint=disable. Should that case wake the thread anyway? Fixed to not wake up kthread for this case in v3. > > I'm fine to drop my patch if you take this in your series. Thanks a lot for your review, I prefer to keep the one in my series, I feel it will be more flexible to control writeback via the new sysfs nodes. > > Thanks, > Daeho > >> >>> >>> Fixes: 7a1cf2a76b71 ("f2fs: cache: use meta cache") >>> Fixes: 493b9dc8b52c ("f2fs: cache: use node cache") >>> Signed-off-by: Daeho Jeong >>> --- >>> fs/f2fs/cache.c | 59 +++++++++++++++++++++++++++++++++++++++++++++-- >>> fs/f2fs/cache.h | 1 + >>> fs/f2fs/segment.h | 13 +++++++---- >>> 3 files changed, 67 insertions(+), 6 deletions(-) >>> >>> diff --git a/fs/f2fs/cache.c b/fs/f2fs/cache.c >>> index 04b5408cbc5..6b6f49fae25 100644 >>> --- a/fs/f2fs/cache.c >>> +++ b/fs/f2fs/cache.c >>> @@ -61,6 +61,42 @@ void f2fs_cache_update_tag(struct f2fs_cached_block *entry, >>> spin_unlock_irqrestore(&cache->tree_lock, flags); >>> } >>> >>> +/* >>> + * While checkpoint is disabled, checkpoint does not write back dirty node and >>> + * meta caches, and the writeback thread is the only one that writes them. Tell >>> + * whether there are as many dirty caches as f2fs_write_node_caches() and >>> + * f2fs_write_meta_caches() collect before writing, so that the thread should >>> + * write them now instead of every cache_wb_interval. >>> + */ >>> +static bool f2fs_cache_wb_needed(struct f2fs_sb_info *sbi) >>> +{ >>> + if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) >>> + return false; >>> + >>> + return get_nr_caches(sbi, F2FS_DIRTY_NODES) >= >>> + nr_caches_to_collect(sbi, NODE) || >>> + get_nr_caches(sbi, F2FS_DIRTY_META) >= >>> + nr_caches_to_collect(sbi, META); >>> +} >>> + >>> +static void f2fs_wake_cache_wb_thread(struct f2fs_sb_info *sbi) >>> +{ >>> + struct f2fs_cache_kthread *cache_thread = &sbi->cache_thread; >>> + >>> + if (!f2fs_cache_wb_needed(sbi)) >>> + return; >>> + >>> + /* pairs with smp_store_release() in f2fs_start_cache_wb_thread() */ >>> + if (!smp_load_acquire(&cache_thread->cache_wb_task)) >>> + return; >>> + >>> + if (READ_ONCE(cache_thread->cache_wb_urgent)) >>> + return; >>> + >>> + WRITE_ONCE(cache_thread->cache_wb_urgent, true); >>> + wake_up(&cache_thread->cache_wb_wq); >>> +} >>> + >>> bool f2fs_mark_cache_dirty(struct f2fs_cached_block *entry) >>> { >>> struct f2fs_cached_block_list *cache = entry->cache; >>> @@ -81,6 +117,7 @@ bool f2fs_mark_cache_dirty(struct f2fs_cached_block *entry) >>> f2fs_cache_update_tag(entry, F2FS_CACHE_TAG_NONE, >>> F2FS_CACHE_TAG_DIRTY); >>> inc_cache_count(cache->sbi, type); >>> + f2fs_wake_cache_wb_thread(cache->sbi); >>> return true; >>> } >>> >>> @@ -679,10 +716,13 @@ static int f2fs_cache_writeback_kthread(void *data) >>> >>> while (!kthread_should_stop()) { >>> unsigned int interval = cache_thread->cache_wb_interval; >>> + s64 nr_dirty; >>> >>> wait_event_freezable_timeout(*wq, >>> - kthread_should_stop(), >>> + kthread_should_stop() || >>> + READ_ONCE(cache_thread->cache_wb_urgent), >>> msecs_to_jiffies(interval)); >>> + WRITE_ONCE(cache_thread->cache_wb_urgent, false); >>> >>> if (kthread_should_stop()) >>> break; >>> @@ -699,10 +739,23 @@ static int f2fs_cache_writeback_kthread(void *data) >>> if (!sb_start_write_trylock(sbi->sb)) >>> continue; >>> >>> + nr_dirty = get_nr_caches(sbi, F2FS_DIRTY_NODES) + >>> + get_nr_caches(sbi, F2FS_DIRTY_META); >>> + >>> f2fs_write_meta_caches(sbi); >>> f2fs_write_node_caches(sbi); >>> >>> sb_end_write(sbi->sb); >>> + >>> + /* >>> + * Each round writes a limited number of caches. Go on without >>> + * waiting while there are still many dirty caches, as long as >>> + * this round wrote some of them. >>> + */ >>> + if (f2fs_cache_wb_needed(sbi) && >>> + get_nr_caches(sbi, F2FS_DIRTY_NODES) + >>> + get_nr_caches(sbi, F2FS_DIRTY_META) < nr_dirty) >>> + WRITE_ONCE(cache_thread->cache_wb_urgent, true); >>> } >>> return 0; >>> } >>> @@ -719,6 +772,7 @@ int f2fs_start_cache_wb_thread(struct f2fs_sb_info *sbi) >>> >>> init_waitqueue_head(&cache_thread->cache_wb_wq); >>> cache_thread->cache_wb_interval = DEF_DIRTY_CACHE_TIMEOUT; >>> + cache_thread->cache_wb_urgent = false; >>> snprintf(name, sizeof(name), "f2fs_writeback-%u:%u", >>> MAJOR(dev), MINOR(dev)); >>> >>> @@ -726,7 +780,8 @@ int f2fs_start_cache_wb_thread(struct f2fs_sb_info *sbi) >>> if (IS_ERR(task)) >>> return PTR_ERR(task); >>> >>> - cache_thread->cache_wb_task = task; >>> + /* pairs with smp_load_acquire() in f2fs_wake_cache_wb_thread() */ >>> + smp_store_release(&cache_thread->cache_wb_task, task); >>> return 0; >>> } >>> >>> diff --git a/fs/f2fs/cache.h b/fs/f2fs/cache.h >>> index 6c4db910d76..1d0216d2128 100644 >>> --- a/fs/f2fs/cache.h >>> +++ b/fs/f2fs/cache.h >>> @@ -232,6 +232,7 @@ struct f2fs_cache_kthread { >>> struct task_struct *cache_wb_task; >>> wait_queue_head_t cache_wb_wq; >>> unsigned int cache_wb_interval; >>> + bool cache_wb_urgent; /* write back without waiting */ >>> }; >>> >>> int f2fs_start_cache_wb_thread(struct f2fs_sb_info *sbi); >>> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h >>> index 526764ba31a..58804e9cc29 100644 >>> --- a/fs/f2fs/segment.h >>> +++ b/fs/f2fs/segment.h >>> @@ -983,11 +983,8 @@ static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) >>> * 512 blocks (2MB) * 8 for nodes, and >>> * 256 blocks * 8 for meta are set. >>> */ >>> -static inline int nr_caches_to_skip(struct f2fs_sb_info *sbi, int type) >>> +static inline int nr_caches_to_collect(struct f2fs_sb_info *sbi, int type) >>> { >>> - if (bdi_wb_dirty_exceeded(sbi->sb->s_bdi)) >>> - return 0; >>> - >>> if (type == DATA) >>> return BLKS_PER_SEG(sbi); >>> else if (type == NODE) >>> @@ -998,6 +995,14 @@ static inline int nr_caches_to_skip(struct f2fs_sb_info *sbi, int type) >>> return 0; >>> } >>> >>> +static inline int nr_caches_to_skip(struct f2fs_sb_info *sbi, int type) >>> +{ >>> + if (bdi_wb_dirty_exceeded(sbi->sb->s_bdi)) >>> + return 0; >>> + >>> + return nr_caches_to_collect(sbi, type); >>> +} >>> + >>> /* >>> * When writing cache asynchronously, align nr_to_write to BIO_MAX_VECS. >>> */ >>