Re: [PATCH 1/2] mm: vmscan: remove folio_test_private() check in pageout()

[Hugh Dickins <hughd@google.com>]

On Sat, 13 Sep 2025, Baolin Wang wrote:
> On 2025/9/13 00:13, Shakeel Butt wrote:
> > On Fri, Sep 12, 2025 at 11:45:07AM +0800, Baolin Wang wrote:
> >> Currently, we no longer attempt to write back filesystem folios in
> >> pageout(),
> >> and only tmpfs/shmem folios and anonymous swapcache folios can be written
> >> back.
> >> Moreover, tmpfs/shmem and swapcache folios do not use the PG_private flag,
> >> which means no fs-private private data is used. Therefore, we can remove
> >> the
> >> redundant folio_test_private() checks and related buffer_head release
> >> logic.
> >>
> >> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> >> ---
> >>   mm/vmscan.c | 16 +---------------
> >>   1 file changed, 1 insertion(+), 15 deletions(-)
> >>
> >> diff --git a/mm/vmscan.c b/mm/vmscan.c
> >> index f1fc36729ddd..8056fccb9cc4 100644
> >> --- a/mm/vmscan.c
> >> +++ b/mm/vmscan.c
> >> @@ -697,22 +697,8 @@ static pageout_t pageout(struct folio *folio, struct
> >> address_space *mapping,
> >>     * swap_backing_dev_info is bust: it doesn't reflect the
> >>     * congestion state of the swapdevs.  Easy to fix, if needed.
> >>     */
> >> -	if (!is_page_cache_freeable(folio))
> >> +	if (!is_page_cache_freeable(folio) || !mapping)
> >>   		return PAGE_KEEP;
> >> -	if (!mapping) {
> >> -		/*
> >> -		 * Some data journaling orphaned folios can have
> >> -		 * folio->mapping == NULL while being dirty with clean
> >> buffers.
> >> -		 */
> > 
> > Can this case not happen anymore and try_to_free_buffers is not needed?
> 
> For dirty file folios, pageout() will return PAGE_KEEP and put them back on
> the LRU list. So even if mapping = NULL, background workers for writeback will
> continue to handle them, rather than in shrink_folio_list().

You've persuaded everyone else, but I'm still not convinced:
what are those "background workers for writeback",
that manage to work on orphaned folios with NULL mapping?

I think *this* is the place which deals with that case, and you're
now proposing to remove it (and returning PAGE_KEEP not PAGE_CLEAN,
so it misses the filemap_release_folio() below the switch(pageout())).

There's even a comment over in migrate_folio_unmap():
"Everywhere else except page reclaim, the page is invisible to the vm".

And your argument that the code *afterwards* rejects everything but
shmem or anon, and neither of those would have folio_test_private(),
certainly did not convince me.

Please persuade me.  But I've no evidence that this case does or does
not still arise; and agree that there must be cleaner ways of doing it.

Hugh

> >> -		if (folio_test_private(folio)) {
> >> -			if (try_to_free_buffers(folio)) {
> >> -				folio_clear_dirty(folio);
> >> -				pr_info("%s: orphaned folio\n", __func__);
> >> -				return PAGE_CLEAN;
> >> -			}
> >> -		}
> >> -		return PAGE_KEEP;
> >> -	}
> >>   
> >>    if (!shmem_mapping(mapping) && !folio_test_anon(folio))
> >>   		return PAGE_ACTIVATE;
> >> -- 
> >> 2.43.7