Re: make alloc_anon_inode more useful

[Christoph Hellwig <hch@lst.de>]

On Wed, Mar 10, 2021 at 04:05:45AM +0000, Matthew Wilcox wrote:
> On Tue, Mar 09, 2021 at 04:53:39PM +0100, Christoph Hellwig wrote:
> > this series first renames the existing alloc_anon_inode to
> > alloc_anon_inode_sb to clearly mark it as requiring a superblock.
> > 
> > It then adds a new alloc_anon_inode that works on the anon_inode
> > file system super block, thus removing tons of boilerplate code.
> > 
> > The few remainig callers of alloc_anon_inode_sb all use alloc_file_pseudo
> > later, but might also be ripe for some cleanup.
> 
> On a somewhat related note, could I get you to look at
> drivers/video/fbdev/core/fb_defio.c?
> 
> As far as I can tell, there's no need for fb_deferred_io_aops to exist.
> We could just set file->f_mapping->a_ops to NULL, and set_page_dirty()
> would do the exact same thing this code does (except it would get the
> return value correct).

> But maybe that would make something else go wrong that distinguishes
> between page->mapping being NULL and page->mapping->a_ops->foo being NULL?

I can't find any place in the kernel that treats a NULL aops different
from not having the method it is looking for. 

> Completely untested patch ...

the patch looks mostly good to me.

>  	}
> -#ifdef CONFIG_FB_DEFERRED_IO
> -	if (info->fbdefio)
> -		fb_deferred_io_open(info, inode, file);
> -#endif
> +	file->f_mapping->a_ops = NULL;

But I'd also skip this.  Drivers generally do not set aops, but if they
do a funtion like this really should not override it.  This will require
an audit of the callers, though.