On Mon, Feb 06, 2023 at 05:04:07PM -0800, Andrew Morton wrote:
> +struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
> + pgoff_t index, gfp_t gfp)
> +{
> + struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
> + struct page *page = folio_file_page(folio, index);
> +
Ugh, insufficiently cautious.
This should have been
struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
struct page *page;
if (IS_ERR(folio))
return &folio->page;
page = folio_file_page(folio, index);
Unfortunately I have to go out now, but that's the solution to the
latest syzbot splat I saw. I'll do you a proper -fix patch later
if you don't get to it first.
> if (PageHWPoison(page)) {
> folio_put(folio);
> return ERR_PTR(-EIO);
> }
>
> return page;
> -#else
> - /*
> - * The tiny !SHMEM case uses ramfs without swap
> - */
> - return read_cache_page_gfp(mapping, index, gfp);
> -#endif
> }
> EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
> _
>
> Patches currently in -mm which might be from willy@xxxxxxxxxxxxx are
>
> mm-add-memcpy_from_file_folio.patch
> filemap-add-mapping_read_folio_gfp.patch
> shmem-add-shmem_read_folio-and-shmem_read_folio_gfp.patch
> shmem-fix-w=1-build-warnings-with-config_shmem=n.patch
>
