On Sat, Jun 03, 2023 at 07:21:01PM -0700, Eric Biggers wrote:
> bio_first_page_all(bio)->mapping->host is not compatible with large
> folios, since the first page of the bio is not necessarily the head page
> of the folio, and therefore it might not have the mapping pointer set.
Yes, that's true. It is going to depend on the filesystem, since these
two bios are equivalent:
(folio->page[0], offset=0x4000, len=0x300)
(folio->page[4], offset=0, len=0x300)
and we don't yet have a rule that filesystems must construct one or the
other. We probably _should_, but that was pretty low down my list of
things to care about right now.
> Therefore, move the dereference of ->mapping->host into
> verify_data_blocks(), which works with a folio.
Seems reasonable.
Reviewed-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>
> @@ -320,7 +321,6 @@ EXPORT_SYMBOL_GPL(fsverity_verify_blocks);
> */
> void fsverity_verify_bio(struct bio *bio)
> {
> - struct inode *inode = bio_first_page_all(bio)->mapping->host;
An alternative fix could be
struct folio *first = page_folio(bio_first_page_all(bio));
struct inode *inode = first->mapping->host;
Or we could add a bio_first_folio_all() that wraps that for you.
