[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <da1b15f0-95bd-43ab-9900-c3afd87e9ac7@kernel.org>
Date: Thu, 7 Aug 2025 10:47:43 +0800
From: Chao Yu <chao@...nel.org>
To: Zhiguo Niu <niuzhiguo84@...il.com>
Cc: chao@...nel.org, jaegeuk@...nel.org, linux-kernel@...r.kernel.org,
linux-f2fs-devel@...ts.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH] f2fs: introduce flush_policy sysfs entry
On 8/7/25 10:16, Zhiguo Niu wrote:
> Chao Yu via Linux-f2fs-devel <linux-f2fs-devel@...ts.sourceforge.net>
> 于2025年7月31日周四 16:49写道:
>>
>> This patch introduces a new sysfs entry /sys/fs/f2fs/<disk>/flush_policy
>> in order to tune performance of f2fs data flush flow.
>>
>> For example, checkpoint will use REQ_FUA to persist CP metadata, however,
>> some kind device has bad performance on REQ_FUA command, result in that
>> checkpoint being blocked for long time, w/ this sysfs entry, we can give
>> an option to use REQ_PREFLUSH command instead of REQ_FUA during checkpoint,
>> it can help to mitigate long latency of checkpoint.
>>
>> Signed-off-by: Chao Yu <chao@...nel.org>
>> ---
>> Documentation/ABI/testing/sysfs-fs-f2fs | 9 +++++++++
>> fs/f2fs/checkpoint.c | 10 +++++++++-
>> fs/f2fs/f2fs.h | 7 +++++++
>> fs/f2fs/sysfs.c | 9 +++++++++
>> 4 files changed, 34 insertions(+), 1 deletion(-)
>>
>> diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
>> index bc0e7fefc39d..2fedb44b713b 100644
>> --- a/Documentation/ABI/testing/sysfs-fs-f2fs
>> +++ b/Documentation/ABI/testing/sysfs-fs-f2fs
>> @@ -883,3 +883,12 @@ Date: June 2025
>> Contact: "Daeho Jeong" <daehojeong@...gle.com>
>> Description: Control GC algorithm for boost GC. 0: cost benefit, 1: greedy
>> Default: 1
>> +
>> +What: /sys/fs/f2fs/<disk>/flush_policy
>> +Date: July 2025
>> +Contact: "Chao Yu" <chao@...nel.org>
>> +Description: Device has different performance for the same flush methods, this node
>> + can be used to tune performance by setting different flush methods.
>> +
>> + policy value description
>> + 0x00000001 Use preflush instead of fua during checkpoint
>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
>> index bbe07e3a6c75..10cc27158770 100644
>> --- a/fs/f2fs/checkpoint.c
>> +++ b/fs/f2fs/checkpoint.c
>> @@ -1419,7 +1419,9 @@ static void commit_checkpoint(struct f2fs_sb_info *sbi,
>> f2fs_folio_put(folio, false);
>>
>> /* submit checkpoint (with barrier if NOBARRIER is not set) */
>> - f2fs_submit_merged_write(sbi, META_FLUSH);
>> + f2fs_submit_merged_write(sbi,
>> + sbi->flush_policy & BIT(FLUSH_POLICY_CP_NO_FUA) ?
>> + META : META_FLUSH);
>> }
>>
>> static inline u64 get_sectors_written(struct block_device *bdev)
>> @@ -1606,6 +1608,12 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
>> f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
>> stat_cp_time(cpc, CP_TIME_WAIT_LAST_CP);
>>
>> + if (sbi->flush_policy & BIT(FLUSH_POLICY_CP_NO_FUA)) {
>> + err = f2fs_flush_device_cache(sbi);
>> + if (err)
>> + return err;
>> + }
> Hi Chao,
> sorry for late interruption.
> f2fs_flush_device_cache is just used for multi device cases, so this patch will
> not be suitable for a single device if the user changes flush_policy
> by sysfs, otherwise it
> will not use FUA write when committing checkpoint? Is this expected?
Zhiguo, I expected to flush primary device as well, so I need to fix this, thanks
for catching this.
Thanks,
> Thanks!
>> +
>> /*
>> * invalidate intermediate page cache borrowed from meta inode which are
>> * used for migration of encrypted, verity or compressed inode's blocks.
>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>> index 5c0ea60e502e..5bba24ca15c8 100644
>> --- a/fs/f2fs/f2fs.h
>> +++ b/fs/f2fs/f2fs.h
>> @@ -1621,6 +1621,11 @@ struct decompress_io_ctx {
>> #define MAX_COMPRESS_LOG_SIZE 8
>> #define MAX_COMPRESS_WINDOW_SIZE(log_size) ((PAGE_SIZE) << (log_size))
>>
>> +enum flush_policy {
>> + FLUSH_POLICY_CP_NO_FUA,
>> + FLUSH_POLICY_MAX,
>> +};
>> +
>> struct f2fs_sb_info {
>> struct super_block *sb; /* pointer to VFS super block */
>> struct proc_dir_entry *s_proc; /* proc entry */
>> @@ -1873,6 +1878,8 @@ struct f2fs_sb_info {
>> /* carve out reserved_blocks from total blocks */
>> bool carve_out;
>>
>> + unsigned int flush_policy; /* flush policy */
>> +
>> #ifdef CONFIG_F2FS_FS_COMPRESSION
>> struct kmem_cache *page_array_slab; /* page array entry */
>> unsigned int page_array_slab_size; /* default page array slab size */
>> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
>> index f736052dea50..b69015f1dc67 100644
>> --- a/fs/f2fs/sysfs.c
>> +++ b/fs/f2fs/sysfs.c
>> @@ -852,6 +852,13 @@ static ssize_t __sbi_store(struct f2fs_attr *a,
>> return count;
>> }
>>
>> + if (!strcmp(a->attr.name, "flush_policy")) {
>> + if (t >= BIT(FLUSH_POLICY_MAX))
>> + return -EINVAL;
>> + *ui = (unsigned int)t;
>> + return count;
>> + }
>> +
>> if (!strcmp(a->attr.name, "gc_boost_gc_multiple")) {
>> if (t < 1 || t > SEGS_PER_SEC(sbi))
>> return -EINVAL;
>> @@ -1175,6 +1182,7 @@ F2FS_SBI_GENERAL_RW_ATTR(blkzone_alloc_policy);
>> #endif
>> F2FS_SBI_GENERAL_RW_ATTR(carve_out);
>> F2FS_SBI_GENERAL_RW_ATTR(reserved_pin_section);
>> +F2FS_SBI_GENERAL_RW_ATTR(flush_policy);
>>
>> /* STAT_INFO ATTR */
>> #ifdef CONFIG_F2FS_STAT_FS
>> @@ -1371,6 +1379,7 @@ static struct attribute *f2fs_attrs[] = {
>> ATTR_LIST(max_read_extent_count),
>> ATTR_LIST(carve_out),
>> ATTR_LIST(reserved_pin_section),
>> + ATTR_LIST(flush_policy),
>> NULL,
>> };
>> ATTRIBUTE_GROUPS(f2fs);
>> --
>> 2.49.0
>>
>>
>>
>> _______________________________________________
>> Linux-f2fs-devel mailing list
>> Linux-f2fs-devel@...ts.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
Powered by blists - more mailing lists