On Wed, Mar 23, 2022 at 3:42 PM Christoph Hellwig <hch@xxxxxx> wrote:
>
> On Wed, Mar 23, 2022 at 06:38:22AM +0900, Ryusuke Konishi wrote:
> > This looks because the mask of GFP_KERNEL is removed along with
> > the removal of mpage_alloc().
> >
>
> > The default value of the gfp flag is set to GFP_HIGHUSER_MOVABLE by
> > inode_init_always().
> > So, __GFP_HIGHMEM hits the gfp warning at bio_alloc() that
> > do_mpage_readpage() calls.
>
> Yeah. Let's try this to match the iomap code:
>
> diff --git a/fs/mpage.c b/fs/mpage.c
> index 9ed1e58e8d70b..d465883edf719 100644
> --- a/fs/mpage.c
> +++ b/fs/mpage.c
> @@ -148,13 +148,11 @@ static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
> int op = REQ_OP_READ;
> unsigned nblocks;
> unsigned relative_block;
> - gfp_t gfp;
> + gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
>
> if (args->is_readahead) {
> op |= REQ_RAHEAD;
> - gfp = readahead_gfp_mask(page->mapping);
> - } else {
> - gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
> + gfp |= __GFP_NORETRY | __GFP_NOWARN;
> }
>
> if (page_has_buffers(page))
I did not test for iomap, but this patch has fixed the same regression on the
latest mainline at least for ext2, exfat, vfat and nilfs2. Thanks!
Ryusuke Konishi
