[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20160316142729.GA125481@black.fi.intel.com>
Date: Wed, 16 Mar 2016 17:27:29 +0300
From: "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
To: "Kirill A. Shutemov" <kirill@...temov.name>
Cc: Richard Weinberger <richard@....at>,
Christoph Hellwig <hch@...radead.org>,
linux-fsdevel <linux-fsdevel@...r.kernel.org>,
"linux-mtd@...ts.infradead.org" <linux-mtd@...ts.infradead.org>,
"linux-mm@...ck.org" <linux-mm@...ck.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Boris Brezillon <boris.brezillon@...e-electrons.com>,
Maxime Ripard <maxime.ripard@...e-electrons.com>,
David Gstir <david@...ma-star.at>,
Dave Chinner <david@...morbit.com>,
Artem Bityutskiy <dedekind1@...il.com>,
Alexander Kaplan <alex@...tthing.co>
Subject: Re: Page migration issue with UBIFS
On Wed, Mar 16, 2016 at 05:21:56PM +0300, Kirill A. Shutemov wrote:
> On Wed, Mar 16, 2016 at 12:18:50AM +0100, Richard Weinberger wrote:
> > Am 15.03.2016 um 16:37 schrieb Christoph Hellwig:
> > > On Tue, Mar 15, 2016 at 04:32:40PM +0100, Richard Weinberger wrote:
> > >>> Or if ->page_mkwrite() was called, why the page is not dirty?
> > >>
> > >> BTW: UBIFS does not implement ->migratepage(), could this be a problem?
> > >
> > > This might be the reason. I can't reall make sense of
> > > buffer_migrate_page, but it seems to migrate buffer_head state to
> > > the new page.
> > >
> > > I'd love to know why CMA even tries to migrate pages that don't have a
> > > ->migratepage method, this seems incredibly dangerous to me.
> >
> > FYI, with a dummy ->migratepage() which returns only -EINVAL UBIFS does no
> > longer explode upon page migration.
> > Tomorrow I'll do more tests to make sure.
>
> Could you check if something like this would fix the issue.
> Completely untested.
>
> diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c
> index 065c88f8e4b8..9da34120dc5e 100644
> --- a/fs/ubifs/file.c
> +++ b/fs/ubifs/file.c
> @@ -52,6 +52,7 @@
> #include "ubifs.h"
> #include <linux/mount.h>
> #include <linux/slab.h>
> +#include <linux/migrate.h>
>
> static int read_block(struct inode *inode, void *addr, unsigned int block,
> struct ubifs_data_node *dn)
> @@ -1452,6 +1453,20 @@ static int ubifs_set_page_dirty(struct page *page)
> return ret;
> }
>
> +static int ubifs_migrate_page(struct address_space *mapping,
> + struct page *newpage, struct page *page, enum migrate_mode mode)
> +{
> + if (PagePrivate(page)) {
> + SetPagePrivate(newpage);
> + __set_page_dirty_nobuffers(newpage);
> + }
> +
> + if (PageChecked(page))
> + SetPageChecked(newpage);
These two lines are redundant, migrate_page_copy() would do this for us.
> +
> + return migrate_page(mapping, newpage, page, mode);
> +}
> +
> static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
> {
> /*
> @@ -1591,6 +1606,7 @@ const struct address_space_operations ubifs_file_address_operations = {
> .write_end = ubifs_write_end,
> .invalidatepage = ubifs_invalidatepage,
> .set_page_dirty = ubifs_set_page_dirty,
> + .migratepage = ubifs_migrate_page,
> .releasepage = ubifs_releasepage,
> };
>
> --
> Kirill A. Shutemov
--
Kirill A. Shutemov
Powered by blists - more mailing lists