Re: [f2fs-dev] [PATCH] f2fs: avoid frequent costly fsck triggers

From: Chao Yu
Date: Fri Dec 07 2018 - 04:41:24 EST


On 2018/12/1 4:28, Jaegeuk Kim wrote:
> On 11/30, Chao Yu wrote:
>> On 2018/11/30 10:35, Sheng Yong wrote:
>>> Hi, Jaegeuk and Chao,
>>>
>>> On 2018/11/29 1:48, Jaegeuk Kim wrote:
>>>> On 11/28, Chao Yu wrote:
>>>>> On 2018/11/28 16:10, Jaegeuk Kim wrote:
>>>>>> On 11/28, Chao Yu wrote:
>>>>>>> Hi Jaeguek,
>>>>>>>
>>>>>>> On 2018/11/28 15:31, Jaegeuk Kim wrote:
>>>>>>>> If we want to re-enable nat_bits, we rely on fsck which requires full scan
>>>>>>>> of directory tree. Let's do that by regular fsck or unclean shutdown.
>>>>>>>
>>>>>>> Reviewed-by: Chao Yu <yuchao0@xxxxxxxxxx>
>>>>>>>
>>>>>>> BTW, I have patch made some month ago...
>>>>>>>
>>>>>>> In order to detect nat_bits disabling, could we introduce one more flag for
>>>>>>> fsck?
>>>>>>
>>>>>> Do we have a way to enable nat_bits very quickly in fsck?
>>>>>
>>>>> For image with SBI_NATBIT_NEED_REPAIR flag, can we just check metadata and
>>>>> rebuild nat_bits based on verified nat blocks/journals?
>>>>
>>>> I'm leaning to rely on full scan to enable nat_bits again. We may add a mount
>>>> count or timer to trigger fsck regularly?
>>>
>>> I'm afraid regular full fsck would give us bad experience of booting time.
>>> FYI, 256GB storage, which is filled with small files, costs almost 10 min
>>> to do a full fsck. And it seems larger capacity storages are on the way.
>>
>> Agreed.
>
> Agreed. So, that's why I wrote this patch.

I see.

>
>>
>>> So, is it worth doing that only to enable nat_bits (plus checking f2fs
>>> consistent not that necessarily)?
>>
>> In android environment, I think it may be too expensive for adding nat_bits
>> by triggering full scan by fsck during boot time.
>
> That's why I'd like to enable this only when we need full scan.
>
>>
>> If we can update all nat bitmap in free time after mount, maybe we can
>> rebuild nat_bits based on full nat bitmap during umount, which can be
>> cheaper than rebuiding in userspace.
>
> Yeah, rebuiling nat_bits in run time would be better, but can be applied
> in future. But, since Android reboot procedure uses a timeout, if we exceed

It only need to write additional nat_bits blocks during last umount
checkpoint, why it can exceed the timeout period?

Thanks,

> it, we'll get unclean unmount which triggers another fsck, which doesn't
> make sense at all.
>
>>
>> Thanks,
>>
>>>
>>> Thanks
>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>>>
>>>>>>>
>>>>>>> >From 86e8bdb2faeec904944bb6621073f4f7de51cc2d Mon Sep 17 00:00:00 2001
>>>>>>> From: Chao Yu <yuchao0@xxxxxxxxxx>
>>>>>>> Date: Sun, 9 Sep 2018 05:40:20 +0800
>>>>>>> Subject: [PATCH] f2fs: set specified flag after invalidating nat_bits
>>>>>>>
>>>>>>> Signed-off-by: Chao Yu <yuchao0@xxxxxxxxxx>
>>>>>>> ---
>>>>>>> fs/f2fs/checkpoint.c | 12 +++++++++++-
>>>>>>> fs/f2fs/f2fs.h | 3 ++-
>>>>>>> fs/f2fs/node.c | 3 +++
>>>>>>> include/linux/f2fs_fs.h | 1 +
>>>>>>> 4 files changed, 17 insertions(+), 2 deletions(-)
>>>>>>>
>>>>>>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
>>>>>>> index 7e17bb3dfcb1..f7fb14e0f5f9 100644
>>>>>>> --- a/fs/f2fs/checkpoint.c
>>>>>>> +++ b/fs/f2fs/checkpoint.c
>>>>>>> @@ -1226,13 +1226,16 @@ static void update_ckpt_flags(struct f2fs_sb_info
>>>>>>> *sbi, struct cp_control *cpc)
>>>>>>> unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
>>>>>>> struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
>>>>>>> unsigned long flags;
>>>>>>> + bool disable_natbits = false;
>>>>>>>
>>>>>>> spin_lock_irqsave(&sbi->cp_lock, flags);
>>>>>>>
>>>>>>> if ((cpc->reason & CP_UMOUNT) &&
>>>>>>> le32_to_cpu(ckpt->cp_pack_total_block_count) >
>>>>>>> - sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
>>>>>>> + sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks) {
>>>>>>> disable_nat_bits(sbi, false);
>>>>>>> + disable_natbits = true;
>>>>>>> + }
>>>>>>>
>>>>>>> if (cpc->reason & CP_TRIMMED)
>>>>>>> __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
>>>>>>> @@ -1270,11 +1273,18 @@ static void update_ckpt_flags(struct f2fs_sb_info
>>>>>>> *sbi, struct cp_control *cpc)
>>>>>>> if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
>>>>>>> __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
>>>>>>>
>>>>>>> + if (is_sbi_flag_set(sbi, SBI_NATBIT_NEED_REPAIR))
>>>>>>> + __set_ckpt_flags(ckpt, CP_NATBIT_NEED_FSCK_FLAG);
>>>>>>> +
>>>>>>> /* set this flag to activate crc|cp_ver for recovery */
>>>>>>> __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
>>>>>>> __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
>>>>>>>
>>>>>>> spin_unlock_irqrestore(&sbi->cp_lock, flags);
>>>>>>> +
>>>>>>> + if (disable_natbits)
>>>>>>> + f2fs_msg(sbi->sb, KERN_NOTICE,
>>>>>>> + "No enough space in CP area, disable nat_bits.");
>>>>>>> }
>>>>>>>
>>>>>>> static void commit_checkpoint(struct f2fs_sb_info *sbi,
>>>>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>>>>>>> index f0cedbe0c536..b55341c269b2 100644
>>>>>>> --- a/fs/f2fs/f2fs.h
>>>>>>> +++ b/fs/f2fs/f2fs.h
>>>>>>> @@ -1107,6 +1107,7 @@ enum {
>>>>>>> SBI_QUOTA_NEED_FLUSH, /* need to flush quota info in CP */
>>>>>>> SBI_QUOTA_SKIP_FLUSH, /* skip flushing quota in current CP */
>>>>>>> SBI_QUOTA_NEED_REPAIR, /* quota file may be corrupted */
>>>>>>> + SBI_NATBIT_NEED_REPAIR, /* nat full/empty bitmaps need repair */
>>>>>>> };
>>>>>>>
>>>>>>> enum {
>>>>>>> @@ -1628,7 +1629,7 @@ static inline void disable_nat_bits(struct
>>>>>>> f2fs_sb_info *sbi, bool lock)
>>>>>>> {
>>>>>>> unsigned long flags;
>>>>>>>
>>>>>>> - set_sbi_flag(sbi, SBI_NEED_FSCK);
>>>>>>> + set_sbi_flag(sbi, SBI_NATBIT_NEED_REPAIR);
>>>>>>>
>>>>>>> if (lock)
>>>>>>> spin_lock_irqsave(&sbi->cp_lock, flags);
>>>>>>> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
>>>>>>> index e57add1e8966..0c6f8312a087 100644
>>>>>>> --- a/fs/f2fs/node.c
>>>>>>> +++ b/fs/f2fs/node.c
>>>>>>> @@ -2902,6 +2902,9 @@ static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
>>>>>>>
>>>>>>> cp_ver |= (cur_cp_crc(ckpt) << 32);
>>>>>>> if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
>>>>>>> + f2fs_msg(sbi->sb, KERN_NOTICE,
>>>>>>> + "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
>>>>>>> + cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
>>>>>>> disable_nat_bits(sbi, true);
>>>>>>> return 0;
>>>>>>> }
>>>>>>> diff --git a/include/linux/f2fs_fs.h b/include/linux/f2fs_fs.h
>>>>>>> index 7196653833fa..1f3ae1504573 100644
>>>>>>> --- a/include/linux/f2fs_fs.h
>>>>>>> +++ b/include/linux/f2fs_fs.h
>>>>>>> @@ -117,6 +117,7 @@ struct f2fs_super_block {
>>>>>>> /*
>>>>>>> * For checkpoint
>>>>>>> */
>>>>>>> +#define CP_NATBIT_NEED_FSCK_FLAG 0X00002000
>>>>>>> #define CP_DISABLED_FLAG 0x00001000
>>>>>>> #define CP_QUOTA_NEED_FSCK_FLAG 0x00000800
>>>>>>> #define CP_LARGE_NAT_BITMAP_FLAG 0x00000400
>>>>>>> --
>>>>>>> 2.18.0.rc1
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
>>>>>>>> ---
>>>>>>>> fs/f2fs/f2fs.h | 6 +++++-
>>>>>>>> 1 file changed, 5 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>>>>>>>> index c28a9d1cb278..aa500239baf2 100644
>>>>>>>> --- a/fs/f2fs/f2fs.h
>>>>>>>> +++ b/fs/f2fs/f2fs.h
>>>>>>>> @@ -1621,7 +1621,11 @@ static inline void disable_nat_bits(struct f2fs_sb_info *sbi, bool lock)
>>>>>>>> {
>>>>>>>> unsigned long flags;
>>>>>>>>
>>>>>>>> - set_sbi_flag(sbi, SBI_NEED_FSCK);
>>>>>>>> + /*
>>>>>>>> + * In order to re-enable nat_bits we need to call fsck.f2fs by
>>>>>>>> + * set_sbi_flag(sbi, SBI_NEED_FSCK). But it may give huge cost,
>>>>>>>> + * so let's rely on regular fsck or unclean shutdown.
>>>>>>>> + */
>>>>>>>>
>>>>>>>> if (lock)
>>>>>>>> spin_lock_irqsave(&sbi->cp_lock, flags);
>>>>>>>>
>>>>>>
>>>>>> .
>>>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Linux-f2fs-devel mailing list
>>>> Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
>>>> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
>>>>
>>>> .
>>>>
>>>
>>>
>>> .
>>>
>
> .
>