On 2017/2/25 2:12, Jaegeuk Kim wrote:
> On 02/24, Yunlong Song wrote:
>> On 2017/2/24 19:37, Chao Yu wrote:
>>> On 2017/2/24 19:11, Yunlong Song wrote:
>>>> I think we do not need to care about the CP_CRC_RECOVERY_FLAG status of old image, I mean we
>>>> do not need to check the already-been-written node footer in the image, what we care about is the
>>>> on-going-to-write node footer, which is used for recovery.
>>>>
>>>> If CP_CRC_RECOVERY_FLAG is defined, then __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG); is
>>>> executed in each do_checkpoint actually, and CP will have that flag for each on-going-to-write node footer.
>>>> I think the recovery process only needs to use the on-going-to-write node rather than the already-been-written
>>>> node in the old image. The already-been-written node in the old image should not appear in the node
>>>> chain of recovery process, right?
>>> Previously, we changed the disk layout of footer in node block, and then we
>>> applied new verifying approach which has better reliability in order to avoid
>>> chaining garbage node block.
>>>
>>> In order to distinguish old disk layout and the new one, we introduce
>>> CP_CRC_RECOVERY_FLAG, once a CP is triggered, we will tag current CP with the
>>> flag, and use new disk layout and new verifying approach for the following node
>>> block updating flow and abnormal power-cut recovery flow.
>>>
>>> For old image which has no CP_CRC_RECOVERY_FLAG flag been set, f2fs needs to use
>>> old disk layout and old verifying approach during recovery for the
>>> compatibility. So that's why we need to check the flag in CP here.
>>
>> For the old disk layout, because we still use new approach to set CP_CRC_RECOVERY_FLAG in the node footer in each do_checkpoint,
>> then I think f2fs should also use new verifying approach during recovery rather than old verifying approach. What is the problem if
>> we do like this?
>
> This is to handle only one case in which:
>
> 1. uses old kernel without this flag,
> 2. calls fsync and gets sudden power-cut,
> 3. updates new kernel having this flag before mount.
>
> Then, if we do not check this flag at mount time, we will lose the last fsync'ed
> node blocks.
Yup, thanks for the explanation. :)
Thanks,
>
> Thanks,
>
>>
>>> Thanks,
>>>
>>>> On 2017/2/24 18:29, Chao Yu wrote:
>>>>> On 2017/2/24 18:06, Yunlong Song wrote:
>>>>>> No need to check the "if" condition each time, just change it to macro codes.
>>>>> We're going to check flag in CP, not just in code of f2fs.
>>>>>
>>>>> Thanks,
>>>>>
>>>>>> Signed-off-by: Yunlong Song <yunlong.song@xxxxxxxxxx>
>>>>>> ---
>>>>>> fs/f2fs/node.h | 20 ++++++++++----------
>>>>>> fs/f2fs/segment.c | 5 +++--
>>>>>> 2 files changed, 13 insertions(+), 12 deletions(-)
>>>>>>
>>>>>> diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
>>>>>> index 3fc9c4b..3e5a58b 100644
>>>>>> --- a/fs/f2fs/node.h
>>>>>> +++ b/fs/f2fs/node.h
>>>>>> @@ -303,11 +303,11 @@ static inline void fill_node_footer_blkaddr(struct page *page, block_t blkaddr)
>>>>>> size_t crc_offset = le32_to_cpu(ckpt->checksum_offset);
>>>>>> __u64 cp_ver = le64_to_cpu(ckpt->checkpoint_ver);
>>>>>>
>>>>>> - if (__is_set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG)) {
>>>>>> - __u64 crc = le32_to_cpu(*((__le32 *)
>>>>>> - ((unsigned char *)ckpt + crc_offset)));
>>>>>> - cp_ver |= (crc << 32);
>>>>>> - }
>>>>>> +#ifdef CP_CRC_RECOVERY_FLAG
>>>>>> + __u64 crc = le32_to_cpu(*((__le32 *)
>>>>>> + ((unsigned char *)ckpt + crc_offset)));
>>>>>> + cp_ver |= (crc << 32);
>>>>>> +#endif
>>>>>> rn->footer.cp_ver = cpu_to_le64(cp_ver);
>>>>>> rn->footer.next_blkaddr = cpu_to_le32(blkaddr);
>>>>>> }
>>>>>> @@ -318,11 +318,11 @@ static inline bool is_recoverable_dnode(struct page *page)
>>>>>> size_t crc_offset = le32_to_cpu(ckpt->checksum_offset);
>>>>>> __u64 cp_ver = cur_cp_version(ckpt);
>>>>>>
>>>>>> - if (__is_set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG)) {
>>>>>> - __u64 crc = le32_to_cpu(*((__le32 *)
>>>>>> - ((unsigned char *)ckpt + crc_offset)));
>>>>>> - cp_ver |= (crc << 32);
>>>>>> - }
>>>>>> +#ifdef CP_CRC_RECOVERY_FLAG
>>>>>> + __u64 crc = le32_to_cpu(*((__le32 *)
>>>>>> + ((unsigned char *)ckpt + crc_offset)));
>>>>>> + cp_ver |= (crc << 32);
>>>>>> +#endif
>>>>>> return cp_ver == cpver_of_node(page);
>>>>>> }
>>>>>>
>>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>>> index 9eb6d89..6c2e1ee 100644
>>>>>> --- a/fs/f2fs/segment.c
>>>>>> +++ b/fs/f2fs/segment.c
>>>>>> @@ -1573,9 +1573,10 @@ static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
>>>>>> {
>>>>>> if (force)
>>>>>> new_curseg(sbi, type, true);
>>>>>> - else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
>>>>>> - type == CURSEG_WARM_NODE)
>>>>>> +#ifndef CP_CRC_RECOVERY_FLAG
>>>>>> + else if (type == CURSEG_WARM_NODE)
>>>>>> new_curseg(sbi, type, false);
>>>>>> +#endif
>>>>>> else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
>>>>>> change_curseg(sbi, type, true);
>>>>>> else
>>>>>>
>>>>> .
>>>>>
>>>>
>>>
>>> .
>>>
>>
>>
>> --
>> Thanks,
>> Yunlong Song
>>
>
> .
>
