Re: [patch 05/10] vfs: pass data to alloc_inode super operation

[Al Viro <viro@ZenIV.linux.org.uk>]

On Tue, Oct 30, 2012 at 08:51:42PM +0000, Al Viro wrote:
> On Tue, Oct 30, 2012 at 04:14:39PM -0400, Jeff Moyer wrote:
> > This patch passes a data pointer along to the alloc_inode
> > super_operations function.  The value will initially be used by
> > bdev_alloc_inode to allocate the bdev_inode on the same numa
> > node as the device to which it is tied.
> 
> Yecchhh...
> 
> > -static struct inode *bdev_alloc_inode(struct super_block *sb)
> > +static struct inode *bdev_alloc_inode(struct super_block *sb, void *data)
> >  {
> > -	struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
> > +	struct bdev_inode *ei;
> > +	int partno;
> > +	int node;
> > +
> > +	if (data) {
> > +		struct gendisk *disk;
> > +		dev_t dev = *(dev_t *)data;
> > +		disk = get_gendisk(dev, &partno);
> > +		node = disk->node_id;

BTW, speaking of other reasons this is FUBAR - mknod a block device node
for something that doesn't exist and try to open it.  struct block_device
lifetime is not limited to that of struct gendisk; it can be allocated
before gendisk is there.  Moreover, with static /dev and demand-loaded
modules this will be the normal scenario.

IOW, this is completely misguided.