The folio returned from __filemap_get_folio may be a NULL, it will
causes the kernel crash when access folio->page.

Signed-off-by: Tuo Cao <91tuocao@gmail.com>
---
 mm/folio-compat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/folio-compat.c b/mm/folio-compat.c
index e1e23b4947d7..37ba33135506 100644
--- a/mm/folio-compat.c
+++ b/mm/folio-compat.c
@@ -108,7 +108,7 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
 	struct folio *folio;
 
 	folio = __filemap_get_folio(mapping, index, fgp_flags, gfp);
-	if ((fgp_flags & FGP_HEAD) || !folio || xa_is_value(folio))
+	if ((fgp_flags & FGP_HEAD) || xa_is_value(folio))
 		return &folio->page;
 	return folio_file_page(folio, index);
 }
-- 
2.17.1

On Tue, Oct 18, 2022 at 10:36:39PM +0800, Tuo Cao wrote:
> The folio returned from __filemap_get_folio may be a NULL, it will
> causes the kernel crash when access folio->page.

This is not a bug.  &folio->page does not dereference folio but performs
pointer arithmetic.

On Tue, 18 Oct 2022 15:39:39 +0100 Matthew Wilcox <willy@infradead.org> wrote:

> On Tue, Oct 18, 2022 at 10:36:39PM +0800, Tuo Cao wrote:
> > The folio returned from __filemap_get_folio may be a NULL, it will
> > causes the kernel crash when access folio->page.
> 
> This is not a bug.  &folio->page does not dereference folio but performs
> pointer arithmetic.

This is the third attempt to "fix" this.  So far.  I think we should
take the hint and make the check for a null pointer more explicit.

On Tue, Oct 18, 2022 at 01:48:34PM -0700, Andrew Morton wrote:
> On Tue, 18 Oct 2022 15:39:39 +0100 Matthew Wilcox <willy@infradead.org> wrote:
> 
> > On Tue, Oct 18, 2022 at 10:36:39PM +0800, Tuo Cao wrote:
> > > The folio returned from __filemap_get_folio may be a NULL, it will
> > > causes the kernel crash when access folio->page.
> > 
> > This is not a bug.  &folio->page does not dereference folio but performs
> > pointer arithmetic.
> 
> This is the third attempt to "fix" this.  So far.  I think we should
> take the hint and make the check for a null pointer more explicit.

I think it's doing a great job of pointing out how many alleged
kernel programmers don't actually understand C and aren't capable of
searching mailing lists to see if somebody else already sent a patch.
Hopefully somebody's keeping track and is feeding this information into
their hiring algorithms.

The real solution is to convert all the callers of pagecache_get_page()
to use folios, and then we can delete this code.  I have a small series
in progress.  The amount of text savings is immense; half a kilobyte
from one conversion alone.  Our compound_head() has got very bloated.