Re: [PATCH] mm: Use GFP_KERNEL allocation for the page cache in page_cache_read

[Michal Hocko <mhocko@suse.cz>]

On Thu 19-03-15 14:55:58, Michal Hocko wrote:
> On Thu 19-03-15 08:38:35, Neil Brown wrote:
> [...]
> > Nearly half the places in the kernel which call mapping_gfp_mask() remove the
> > __GFP_FS bit.
> > 
> > That suggests to me that it might make sense to have
> >    mapping_gfp_mask_fs()
> > and
> >    mapping_gfp_mask_nofs()
> >
> > and let the presence of __GFP_FS (and __GFP_IO) be determined by the
> > call-site rather than the filesystem.
> 
> Sounds reasonable to me but filesystems tend to use this in a very
> different ways.
> - xfs drops GFP_FS in xfs_setup_inode so all page cache allocations are
>   NOFS.
> - reiserfs drops GFP_FS only before calling read_mapping_page in
>   reiserfs_get_page and never restores the original mask.
> - btrfs doesn't seem to rely on mapping_gfp_mask for anything other than
>   btree_inode (unless it gets inherrited in a way I haven't noticed).
> - ext* doesn't seem to rely on the mapping gfp mask at all.
> 
> So it is not clear to me how we should change that into callsites. But I
> guess we can change at least the page fault path like the following. I
> like it much more than the previous way which is too hackish.

But this is racy instead... And I do not think we can make it raceless
so scratch this and get back to the original approach.
[...]
> +	/*
> +	 * Some filesystems always drop __GFP_FS to prevent from reclaim
> +	 * recursion back to FS code. This is not the case here because
> +	 * we are at the top of the call chain. Add GFP_FS flags to prevent
> +	 * from premature OOM killer.
> +	 */
> +	mapping_gfp = mapping_gfp_mask(mapping);
> +	mapping_set_gfp_mask(mapping, mapping_gfp | __GFP_FS | __GFP_IO);
>  	ret = vma->vm_ops->fault(vma, &vmf);
> +	mapping_set_gfp_mask(mapping, mapping_gfp);
>  	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
>  		return ret;
-- 
Michal Hocko
SUSE Labs