Re: [f2fs-dev] [PATCH] f2fs: skip f2fs_preallocate_blocks() for overwrite case

[Jaegeuk Kim]

On 12/12, Chao Yu wrote:
> Ping,
> 
> On 2021/10/30 11:02, Chao Yu wrote:
> > On 2021/10/30 1:43, Jaegeuk Kim wrote:
> > > On 10/29, Chao Yu wrote:
> > > > Ping,
> > > > 
> > > > On 2021/9/29 8:05, Chao Yu wrote:
> > > > > On 2021/9/29 3:08, Jaegeuk Kim wrote:
> > > > > > On 09/28, Chao Yu wrote:
> > > > > > > In f2fs_file_write_iter(), let's use f2fs_overwrite_io() to
> > > > > > > check whethere it is overwrite case, for such case, we can skip
> > > > > > > f2fs_preallocate_blocks() in order to avoid f2fs_do_map_lock(),
> > > > > > > which may be blocked by checkpoint() potentially.
> > > > > > > 
> > > > > > > Signed-off-by: Chao Yu <chao@xxxxxxxxxx>
> > > > > > > ---
> > > > > > >    fs/f2fs/file.c | 4 ++++
> > > > > > >    1 file changed, 4 insertions(+)
> > > > > > > 
> > > > > > > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > > > > > > index 13deae03df06..51fecb2f4db5 100644
> > > > > > > --- a/fs/f2fs/file.c
> > > > > > > +++ b/fs/f2fs/file.c
> > > > > > > @@ -4321,6 +4321,10 @@ static ssize_t f2fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
> > > > > > >            preallocated = true;
> > > > > > >            target_size = iocb->ki_pos + iov_iter_count(from);
> > > > > > > +        if (f2fs_overwrite_io(inode, iocb->ki_pos,
> > > > > > > +                        iov_iter_count(from)))
> > > > > > > +            goto write;
> > > > > > 
> > > > > > This calls f2fs_map_blocks() which can be duplicate, if it's not the overwirte
> > > > > > case. Do we have other benefit?
> > > > > 
> > > > > f2fs_overwrite_io() will break for append write case w/ below check:
> > > > > 
> > > > >       if (pos + len > i_size_read(inode))
> > > > >           return false;
> > > > > 
> > > > > I guess we may only suffer double f2fs_map_blocks() for write hole
> > > > > case, e.g. truncate to large size & write inside the filesize. For
> > > > > this case, how about adding a condition to allow double f2fs_map_blocks()
> > > > > only if write size is smaller than a threshold?
> > > 
> > > I still don't see the benefit much to do double f2fs_map_blocks. What is the
> > > problem here?
> > 
> > There is potential hangtask happened during swapfile's writeback:
> > 
> > - loop_kthread_worker_fn
> >   - kthread_worker_fn
> >    - loop_queue_work
> >     - lo_rw_aio
> >      - f2fs_file_write_iter
> >       - f2fs_preallocate_blocks
> >        - f2fs_map_blocks
> >         - down_read
> >          - rwsem_down_read_slowpath
> >           - schedule
> > 
> > I try to mitigate such issue by preallocating swapfile's block address and
> > avoid f2fs_do_map_lock() as much as possible in swapfile's writeback path...

How about checking i_blocks and i_size instead of checking the entire map?

> > 
> > Thanks,
> > 
> > > 
> > > > > 
> > > > > Thanks,
> > > > > 
> > > > > > 
> > > > > > > +
> > > > > > >            err = f2fs_preallocate_blocks(iocb, from);
> > > > > > >            if (err) {
> > > > > > >    out_err:
> > > > > > > -- 
> > > > > > > 2.32.0
> > > > > 
> > > > > 
> > > > > _______________________________________________
> > > > > Linux-f2fs-devel mailing list
> > > > > Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > > > > https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.sourceforge.net%2Flists%2Flistinfo%2Flinux-f2fs-devel&amp;data=04%7C01%7Cchao.yu%40oppo.com%7Cbb41006c3f6d4e4d600a08d99b51cbcd%7Cf1905eb1c35341c5951662b4a54b5ee6%7C0%7C0%7C637711597895400286%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=%2BlEAXWLpV5wGX2hL0Wj5p2qX0AqfUFI05Qiqdp8PK8g%3D&amp;reserved=0
> > 
> > 
> > _______________________________________________
> > Linux-f2fs-devel mailing list
> > Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel