Re: [PATCH 1/2] f2fs: remove folio_detach_private() in .invalidate_folio and .release_folio

From: Jaegeuk Kim
Date: Tue Apr 11 2023 - 12:58:29 EST


On 04/11, Chao Yu wrote:
> On 2023/4/11 2:33, Jaegeuk Kim wrote:
> > On 04/10, Chao Yu wrote:
> > > We have maintain PagePrivate and page_private and page reference
> > > w/ {set,clear}_page_private_*, it doesn't need to call
> > > folio_detach_private() in the end of .invalidate_folio and
> > > .release_folio, remove it and use f2fs_bug_on instead.
> > >
> > > Signed-off-by: Chao Yu <chao@xxxxxxxxxx>
> > > ---
> > > fs/f2fs/data.c | 7 +++++--
> > > 1 file changed, 5 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > > index 4946df6dd253..8b179b4bdc03 100644
> > > --- a/fs/f2fs/data.c
> > > +++ b/fs/f2fs/data.c
> > > @@ -3737,7 +3737,8 @@ void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
> > > inode->i_ino == F2FS_COMPRESS_INO(sbi))
> > > clear_page_private_data(&folio->page);
> > > - folio_detach_private(folio);
> > > + f2fs_bug_on(sbi, PagePrivate(&folio->page));
> > > + f2fs_bug_on(sbi, page_private(&folio->page));
> >
> > I think we can just check page_private() only.
>
> Why? how about the case PagePrivate was set, but page_private was't? It must
> be a bug as well?

Given the code, I think both are set all the time. My concern is someone is
not doing set/get properly. Actually, I got a panic on page_private() when
running fsstress overnight. I'm trying to reproduce it to find which bit was
set.

>
> Thanks,
>
> >
> > > }
> > > bool f2fs_release_folio(struct folio *folio, gfp_t wait)
> > > @@ -3759,7 +3760,9 @@ bool f2fs_release_folio(struct folio *folio, gfp_t wait)
> > > clear_page_private_reference(&folio->page);
> > > clear_page_private_gcing(&folio->page);
> > > - folio_detach_private(folio);
> > > + f2fs_bug_on(sbi, PagePrivate(&folio->page));
> > > + f2fs_bug_on(sbi, page_private(&folio->page));
> > > +
> > > return true;
> > > }
> > > --
> > > 2.25.1