On Fri, Jul 10, 2020 at 05:08:04PM +0100, Christoph Hellwig wrote:
> > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
> > index 5daef654956cb..8c3fe7ef56e27 100644
> > --- a/fs/xfs/xfs_icache.c
> > +++ b/fs/xfs/xfs_icache.c
> > @@ -35,15 +35,20 @@ xfs_inode_alloc(
> > xfs_ino_t ino)
> > {
> > struct xfs_inode *ip;
> > + gfp_t gfp_mask = GFP_KERNEL;
> >
> > /*
> > - * if this didn't occur in transactions, we could use
> > - * KM_MAYFAIL and return NULL here on ENOMEM. Set the
> > - * code up to do this anyway.
> > + * If this is inside a transaction, we can not fail here,
> > + * otherwise we can return NULL on ENOMEM.
> > */
> > - ip = kmem_zone_alloc(xfs_inode_zone, 0);
> > +
> > + if (current->flags & PF_MEMALLOC_NOFS)
> > + gfp_mask |= __GFP_NOFAIL;
>
> I'm a little worried about this change in beavior here. Can we
> just keep the unconditional __GFP_NOFAIL and if we really care do the
> change separately after the series? At that point it should probably
> use the re-added PF_FSTRANS flag as well.
Checking PF_FSTRANS was what I suggested should be done here, not
PF_MEMALLOC_NOFS...
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
