On 2020/5/28 10:45, Chao Yu wrote:
> On 2020/5/27 10:20, Sahitya Tummala wrote:
>> In case a compressed file is getting overwritten, the current retry
>> logic doesn't include the current page to be retried now as it sets
>> the new start index as 0 and new end index as writeback_index - 1.
>> This causes the corresponding cluster to be uncompressed and written
>> as normal pages without compression. Fix this by allowing writeback to
>> be retried for the current page as well (in case of compressed page
>> getting retried due to index mismatch with cluster index). So that
>> this cluster can be written compressed in case of overwrite.
>>
>> Signed-off-by: Sahitya Tummala <stumm...@codeaurora.org>
>> ---
>> fs/f2fs/data.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>> index 4af5fcd..bfd1df4 100644
>> --- a/fs/f2fs/data.c
>> +++ b/fs/f2fs/data.c
>> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space
>> *mapping,
>> if ((!cycled && !done) || retry) {
>
> IMO, we add retry logic in wrong place, you can see that cycled value is
> zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
>
> However if retry is true and wbc->range_cyclic is false, then writeback_index
> would be uninitialized variable.
>
> Thoughts?
>
> Thanks,
>
>> cycled = 1;
>> index = 0;
>> - end = writeback_index - 1;
BTW, I notice that range_cyclic writeback flow was refactored in below commit,
and skeleton of f2fs.writepages was copied from
mm/page-writeback.c::write_cache_pages(),
I guess we need follow that change.
64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages
deadlock")
Thanks,
>> + end = retry ? -1 : writeback_index - 1;
>> goto retry;
>> }
>> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
>>
>
>
> _______________________________________________
> Linux-f2fs-devel mailing list
> linux-f2fs-de...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> .
>
