Re: [PATCH] F2FS: invalidate META_MAPPING before IPU/DIO write

[Chao Yu]

On 2021/11/1 15:23, Hyeong-Jun Kim wrote:
> On Mon, 2021-11-01 at 15:12 +0800, Chao Yu wrote:
> > On 2021/11/1 15:09, Hyeong-Jun Kim wrote:
> > > On Mon, 2021-11-01 at 14:28 +0800, Chao Yu wrote:
> > > > On 2021/11/1 13:42, Hyeong-Jun Kim wrote:
> > > > > Encrypted pages during GC are read and cached in META_MAPPING.
> > > > > However, due to cached pages in META_MAPPING, there is an issue
> > > > > where
> > > > > newly written pages are lost by IPU or DIO writes.
> > > > >
> > > > > Thread A                              Thread B
> > > > > - f2fs_gc(): blk 0x10 -> 0x20 (a)
> > > > >                                          - IPU or DIO write on
> > > > > blk
> > > > > 0x20 (b)
> > > > > - f2fs_gc(): blk 0x20 -> 0x30 (c)
> > > > >
> > > > > (a) page for blk 0x20 is cached in META_MAPPING and page for
> > > > > blk
> > > > > 0x10
> > > > >        is invalidated from META_MAPPING.
> > > > > (b) write new data to blk 0x200 using IPU or DIO, but outdated
> > > > > data
> > > > >        still remains in META_MAPPING.
> > > > > (c) f2fs_gc() try to move blk from 0x20 to 0x30 using cached
> > > > > page
> > > > > in
> > > > >        META_MAPPING. In conclusion, the newly written data in
> > > > > (b) is
> > > > > lost.
> > > >
> > > > In c), f2fs_gc() will readahead encrypted block from disk via
> > > > ra_data_block() anyway,
> > > > not matter cached encrypted page of meta inode is uptodate or
> > > > not, so
> > > > it's safe, right?
> > >
> > > Right,
> > > However, if DIO write is performed between phase 3 and phase 4 of
> > > f2fs_gc(),
> > > the cached page of meta_mapping will be out-dated, though it read
> > > data
> > > from
> > > disk via ra_data_block() in phase 3.
> > >
> > > What do you think?
> >
> > Due to i_gc_rwsem lock coverage, the race condition should not happen
> > right now?
> - Thread A                                       - Thread B
> /* phase 3 */
> down_write(i_gc_rwsem)
> ra_data_block()
> up_write(i_gc_rwsem)
>                                                         
>   f2fs_direct_IO() :
>                                                         
>   down_read(i_gc_rwsem)
>                                                         
>   __blockdev_direct_IO()
>                                                             ...
>                                                           
>   get_ddata_block_dio_write()
>                                                             ...
>                                                           
>   f2fs_dio_submit_bio()
>                                                         
>   up_read(i_gc_rwsem)
> /* phase 4 */
> down_write(i_gc_rwsem)
> move_data_block()
> up_write(i_gc_rwsem)
>
> It looks, i_gc_rwsem could not protect page update between phase 3 and
> 4.
>
> Am I missing anything?

It looks you're right, there is a hole in between readahead and movepage functions...

Could you please update the race condition description? and add a tag as below to fix
stable kernel as well:

Fixes: 6aa58d8ad20a ("f2fs: readahead encrypted block during GC")

Thanks,

> Thanks
>
> > Thanks,
> >
> > > Thanks,
> > > > Am I missing anything?
> > > >
> > > > Thanks,
> > > >
> > > > > To address this issue, invalidating pages in META_MAPPING
> > > > > before
> > > > > IPU or
> > > > > DIO write.
> > > > >
> > > > > Signed-off-by: Hyeong-Jun Kim <
> > > > > hj514.kim@xxxxxxxxxxx
> > > > >
> > > > >
> > > > > ---
> > > > >     fs/f2fs/data.c    | 2 ++
> > > > >     fs/f2fs/segment.c | 3 +++
> > > > >     2 files changed, 5 insertions(+)
> > > > >
> > > > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > > > > index 74e1a350c1d8..9f754aaef558 100644
> > > > > --- a/fs/f2fs/data.c
> > > > > +++ b/fs/f2fs/data.c
> > > > > @@ -1708,6 +1708,8 @@ int f2fs_map_blocks(struct inode *inode,
> > > > > struct f2fs_map_blocks *map,
> > > > >     		 */
> > > > >     		f2fs_wait_on_block_writeback_range(inode,
> > > > >     						map->m_pblk,
> > > > > map-
> > > > > > m_len);
> > > > >
> > > > > +		invalidate_mapping_pages(META_MAPPING(sbi),
> > > > > +						map->m_pblk,
> > > > > map-
> > > > > > m_pblk);
> > > > >
> > > > >     
> > > > >     		if (map->m_multidev_dio) {
> > > > >     			block_t blk_addr = map->m_pblk;
> > > > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > > > index 526423fe84ce..f57c55190f9e 100644
> > > > > --- a/fs/f2fs/segment.c
> > > > > +++ b/fs/f2fs/segment.c
> > > > > @@ -3652,6 +3652,9 @@ int f2fs_inplace_write_data(struct
> > > > > f2fs_io_info *fio)
> > > > >     		goto drop_bio;
> > > > >     	}
> > > > >     
> > > > > +	invalidate_mapping_pages(META_MAPPING(fio->sbi),
> > > > > +				fio->new_blkaddr, fio-
> > > > > > new_blkaddr);
> > > > > +
> > > > >     	stat_inc_inplace_blocks(fio->sbi);
> > > > >     
> > > > >     	if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 <<
> > > > > F2FS_IPU_NOCACHE)))