mbox series

[RFC,v2,0/2] Fix gfs2 readahead deadlocks

Message ID 20200703095325.1491832-1-agruenba@redhat.com (mailing list archive)
Headers show
Series Fix gfs2 readahead deadlocks | expand

Message

Andreas Gruenbacher July 3, 2020, 9:53 a.m. UTC
Here's an improved version.  If the IOCB_NOIO flag can be added right
away, we can just fix the locking in gfs2.

Thanks,
Andreas

Andreas Gruenbacher (2):
  fs: Add IOCB_NOIO flag for generic_file_read_iter
  gfs2: Rework read and page fault locking

 fs/gfs2/aops.c     | 45 +--------------------------------------
 fs/gfs2/file.c     | 52 ++++++++++++++++++++++++++++++++++++++++++++--
 include/linux/fs.h |  1 +
 mm/filemap.c       | 17 +++++++++++++--
 4 files changed, 67 insertions(+), 48 deletions(-)

Comments

Linus Torvalds July 3, 2020, 7:24 p.m. UTC | #1
On Fri, Jul 3, 2020 at 2:53 AM Andreas Gruenbacher <agruenba@redhat.com> wrote:
>
> Here's an improved version.  If the IOCB_NOIO flag can be added right
> away, we can just fix the locking in gfs2.

I see nothing wrong with this, and would be ok with getting the
patches as pulls from the gfs2 tree despite touching generic code.

Maybe wait a bit for others to comment (I see Willy already did), but
it seems like a fairly straightforward improvement, and the IOCB_NOIO
flag conceptually seems to match well with the IOCB_NOWAIT one, so
this all makes sense to me.

              Linus
Linus Torvalds July 3, 2020, 7:26 p.m. UTC | #2
On Fri, Jul 3, 2020 at 12:24 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> I see nothing wrong with this [..]

Again, I didn't actually look at the gfs2 parts, but I assume you've
done all the testing of the deadlocks etc.

The IOCB_NOIO patch you can add my acked-by to, fwiw.

          Linus