Re: [f2fs-dev] [PATCH 1/2] f2fs: avoid deadlock caused by lock order of page and lock_op

[Jaegeuk Kim]

Hi Chao,

On 06/26, Chao Yu wrote:
> Hi Jaegeuk,
> 
> On 2017/6/25 0:25, Jaegeuk Kim wrote:
> > - punch_hole
> >  - fill_zero
> >   - f2fs_lock_op
> >   - get_new_data_page
> >    - lock_page
> > 
> > - f2fs_write_data_pages
> >  - lock_page
> >  - do_write_data_page
> >   - f2fs_lock_op
> 
> Good catch!
> 
> With this implementation, page writeback can fail due to concurrent checkpoint,
> this will make fsync/atomic_commit which trigger synchronous write failed randomly.
> 
> How about unifying the lock order in punch_hole as one in writepages for regular
> inode? We can add one more parameter in get_new_data_page to indicate whether
> callee needs to lock cp_rwsem.

Currently, there would be some places to keep cp_rwsem -> page.lock, which seems
not simple to change the lock order with page.lock -> cp_rwsem. IMO, we can retry
flushing data in f2fs_sync_file, once it gets -EAGAIN.

Any thoughts?

> 
> Thanks,
> 
> > 
> > Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
> > ---
> >  fs/f2fs/data.c | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> > 
> > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > index 7d3af48d34a9..9141bd19a902 100644
> > --- a/fs/f2fs/data.c
> > +++ b/fs/f2fs/data.c
> > @@ -1404,8 +1404,9 @@ int do_write_data_page(struct f2fs_io_info *fio)
> >  		}
> >  	}
> >  
> > -	if (fio->need_lock == LOCK_REQ)
> > -		f2fs_lock_op(fio->sbi);
> > +	/* Deadlock due to between page->lock and f2fs_lock_op */
> > +	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
> > +		return -EAGAIN;
> >  
> >  	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
> >  	if (err)
> >